added all files

.dockerignore

@@ -0,0 +1,53 @@
+__pycache__
+.DS_Store
+*.ckpt
+*.safetensors
+*.pth
+*.pt
+*.bin
+*.patch
+*.backup
+*.corrupted
+sorted_styles.json
+/language/default.json
+lena.png
+lena_result.png
+lena_test.py
+config.txt
+config_modification_tutorial.txt
+user_path_config.txt
+user_path_config-deprecated.txt
+build_chb.py
+experiment.py
+/modules/*.png
+/venv
+/tmp
+/ui-config.json
+/outputs
+/config.json
+/log
+/webui.settings.bat
+/embeddings
+/styles.csv
+/params.txt
+/styles.csv.bak
+/webui-user.bat
+/webui-user.sh
+/interrogate
+/user.css
+/.idea
+/notification.ogg
+/notification.mp3
+/SwinIR
+/textual_inversion
+.vscode
+/extensions
+/test/stdout.txt
+/test/stderr.txt
+/cache.json*
+/config_states/
+/node_modules
+/package-lock.json
+/.coverage*
+/auth.json
+*.db

.github/workflows/docker-image.yml

@@ -0,0 +1,32 @@
+name: Docker Image CI
+
+on:
+  push:
+    tags:
+      - v*
+
+jobs:
+
+  build:
+
+    runs-on: ubuntu-latest
+
+    steps:
+      -
+        name: Set up QEMU
+        uses: docker/setup-qemu-action@v3
+      -
+        name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v3
+      -
+        name: Login to Docker Hub
+        uses: docker/login-action@v3
+        with:
+          username: ${{ secrets.DOCKERHUB_USERNAME }}
+          password: ${{ secrets.DOCKERHUB_TOKEN }}
+      -
+        name: Build and push
+        uses: docker/build-push-action@v5
+        with:
+          push: true
+          tags: konieshadow/fooocus-api:latest,konieshadow/fooocus-api:${{ github.ref_name }}

.gitignore

@@ -0,0 +1,50 @@
+#ide config
+.idea
+.vscode
+
+#runtime
+__pycache__
+.DS_Store
+
+# models
+*.ckpt
+*.safetensors
+*.pth
+*.pt
+*.bin
+*.patch
+*.backup
+*.corrupted
+
+# environment
+venv
+.venv
+.env
+.conda
+conda
+
+# log files
+*.log
+logs
+log
+
+# config files
+.cog
+config.txt
+config_modification_tutorial.txt
+user_path_config.txt
+user_path_config-deprecated.txt
+hash_cache.txt
+
+sorted_styles.json
+/presets
+
+# db
+*.db
+
+# cache
+outputs
+
+#other
+*.http
+hash_cache.txt

CreativeMLOpenRAIL-M

@@ -0,0 +1,82 @@
+Copyright (c) 2022 Robin Rombach and Patrick Esser and contributors
+
+CreativeML Open RAIL-M
+dated August 22, 2022
+
+Section I: PREAMBLE
+
+Multimodal generative models are being widely adopted and used, and have the potential to transform the way artists, among other individuals, conceive and benefit from AI or ML technologies as a tool for content creation.
+
+Notwithstanding the current and potential benefits that these artifacts can bring to society at large, there are also concerns about potential misuses of them, either due to their technical limitations or ethical considerations.
+
+In short, this license strives for both the open and responsible downstream use of the accompanying model. When it comes to the open character, we took inspiration from open source permissive licenses regarding the grant of IP rights. Referring to the downstream responsible use, we added use-based restrictions not permitting the use of the Model in very specific scenarios, in order for the licensor to be able to enforce the license in case potential misuses of the Model may occur. At the same time, we strive to promote open and responsible research on generative models for art and content generation.
+
+Even though downstream derivative versions of the model could be released under different licensing terms, the latter will always have to include - at minimum - the same use-based restrictions as the ones in the original license (this license). We believe in the intersection between open and responsible AI development; thus, this License aims to strike a balance between both in order to enable responsible open-science in the field of AI.
+
+This License governs the use of the model (and its derivatives) and is informed by the model card associated with the model.
+
+NOW THEREFORE, You and Licensor agree as follows:
+
+1. Definitions
+
+- "License" means the terms and conditions for use, reproduction, and Distribution as defined in this document.
+- "Data" means a collection of information and/or content extracted from the dataset used with the Model, including to train, pretrain, or otherwise evaluate the Model. The Data is not licensed under this License.
+- "Output" means the results of operating a Model as embodied in informational content resulting therefrom.
+- "Model" means any accompanying machine-learning based assemblies (including checkpoints), consisting of learnt weights, parameters (including optimizer states), corresponding to the model architecture as embodied in the Complementary Material, that have been trained or tuned, in whole or in part on the Data, using the Complementary Material.
+- "Derivatives of the Model" means all modifications to the Model, works based on the Model, or any other model which is created or initialized by transfer of patterns of the weights, parameters, activations or output of the Model, to the other model, in order to cause the other model to perform similarly to the Model, including - but not limited to - distillation methods entailing the use of intermediate data representations or methods based on the generation of synthetic data by the Model for training the other model.
+- "Complementary Material" means the accompanying source code and scripts used to define, run, load, benchmark or evaluate the Model, and used to prepare data for training or evaluation, if any. This includes any accompanying documentation, tutorials, examples, etc, if any.
+- "Distribution" means any transmission, reproduction, publication or other sharing of the Model or Derivatives of the Model to a third party, including providing the Model as a hosted service made available by electronic or other remote means - e.g. API-based or web access.
+- "Licensor" means the copyright owner or entity authorized by the copyright owner that is granting the License, including the persons or entities that may have rights in the Model and/or distributing the Model.
+- "You" (or "Your") means an individual or Legal Entity exercising permissions granted by this License and/or making use of the Model for whichever purpose and in any field of use, including usage of the Model in an end-use application - e.g. chatbot, translator, image generator.
+- "Third Parties" means individuals or legal entities that are not under common control with Licensor or You.
+- "Contribution" means any work of authorship, including the original version of the Model and any modifications or additions to that Model or Derivatives of the Model thereof, that is intentionally submitted to Licensor for inclusion in the Model by the copyright owner or by an individual or Legal Entity authorized to submit on behalf of the copyright owner. For the purposes of this definition, "submitted" means any form of electronic, verbal, or written communication sent to the Licensor or its representatives, including but not limited to communication on electronic mailing lists, source code control systems, and issue tracking systems that are managed by, or on behalf of, the Licensor for the purpose of discussing and improving the Model, but excluding communication that is conspicuously marked or otherwise designated in writing by the copyright owner as "Not a Contribution."
+- "Contributor" means Licensor and any individual or Legal Entity on behalf of whom a Contribution has been received by Licensor and subsequently incorporated within the Model.
+
+Section II: INTELLECTUAL PROPERTY RIGHTS
+
+Both copyright and patent grants apply to the Model, Derivatives of the Model and Complementary Material. The Model and Derivatives of the Model are subject to additional terms as described in Section III.
+
+2. Grant of Copyright License. Subject to the terms and conditions of this License, each Contributor hereby grants to You a perpetual, worldwide, non-exclusive, no-charge, royalty-free, irrevocable copyright license to reproduce, prepare, publicly display, publicly perform, sublicense, and distribute the Complementary Material, the Model, and Derivatives of the Model.
+3. Grant of Patent License. Subject to the terms and conditions of this License and where and as applicable, each Contributor hereby grants to You a perpetual, worldwide, non-exclusive, no-charge, royalty-free, irrevocable (except as stated in this paragraph) patent license to make, have made, use, offer to sell, sell, import, and otherwise transfer the Model and the Complementary Material, where such license applies only to those patent claims licensable by such Contributor that are necessarily infringed by their Contribution(s) alone or by combination of their Contribution(s) with the Model to which such Contribution(s) was submitted. If You institute patent litigation against any entity (including a cross-claim or counterclaim in a lawsuit) alleging that the Model and/or Complementary Material or a Contribution incorporated within the Model and/or Complementary Material constitutes direct or contributory patent infringement, then any patent licenses granted to You under this License for the Model and/or Work shall terminate as of the date such litigation is asserted or filed.
+
+Section III: CONDITIONS OF USAGE, DISTRIBUTION AND REDISTRIBUTION
+
+4. Distribution and Redistribution. You may host for Third Party remote access purposes (e.g. software-as-a-service), reproduce and distribute copies of the Model or Derivatives of the Model thereof in any medium, with or without modifications, provided that You meet the following conditions:
+Use-based restrictions as referenced in paragraph 5 MUST be included as an enforceable provision by You in any type of legal agreement (e.g. a license) governing the use and/or distribution of the Model or Derivatives of the Model, and You shall give notice to subsequent users You Distribute to, that the Model or Derivatives of the Model are subject to paragraph 5. This provision does not apply to the use of Complementary Material.
+You must give any Third Party recipients of the Model or Derivatives of the Model a copy of this License;
+You must cause any modified files to carry prominent notices stating that You changed the files;
+You must retain all copyright, patent, trademark, and attribution notices excluding those notices that do not pertain to any part of the Model, Derivatives of the Model.
+You may add Your own copyright statement to Your modifications and may provide additional or different license terms and conditions - respecting paragraph 4.a. - for use, reproduction, or Distribution of Your modifications, or for any such Derivatives of the Model as a whole, provided Your use, reproduction, and Distribution of the Model otherwise complies with the conditions stated in this License.
+5. Use-based restrictions. The restrictions set forth in Attachment A are considered Use-based restrictions. Therefore You cannot use the Model and the Derivatives of the Model for the specified restricted uses. You may use the Model subject to this License, including only for lawful purposes and in accordance with the License. Use may include creating any content with, finetuning, updating, running, training, evaluating and/or reparametrizing the Model. You shall require all of Your users who use the Model or a Derivative of the Model to comply with the terms of this paragraph (paragraph 5).
+6. The Output You Generate. Except as set forth herein, Licensor claims no rights in the Output You generate using the Model. You are accountable for the Output you generate and its subsequent uses. No use of the output can contravene any provision as stated in the License.
+
+Section IV: OTHER PROVISIONS
+
+7. Updates and Runtime Restrictions. To the maximum extent permitted by law, Licensor reserves the right to restrict (remotely or otherwise) usage of the Model in violation of this License, update the Model through electronic means, or modify the Output of the Model based on updates. You shall undertake reasonable efforts to use the latest version of the Model.
+8. Trademarks and related. Nothing in this License permits You to make use of Licensors’ trademarks, trade names, logos or to otherwise suggest endorsement or misrepresent the relationship between the parties; and any rights not expressly granted herein are reserved by the Licensors.
+9. Disclaimer of Warranty. Unless required by applicable law or agreed to in writing, Licensor provides the Model and the Complementary Material (and each Contributor provides its Contributions) on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied, including, without limitation, any warranties or conditions of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A PARTICULAR PURPOSE. You are solely responsible for determining the appropriateness of using or redistributing the Model, Derivatives of the Model, and the Complementary Material and assume any risks associated with Your exercise of permissions under this License.
+10. Limitation of Liability. In no event and under no legal theory, whether in tort (including negligence), contract, or otherwise, unless required by applicable law (such as deliberate and grossly negligent acts) or agreed to in writing, shall any Contributor be liable to You for damages, including any direct, indirect, special, incidental, or consequential damages of any character arising as a result of this License or out of the use or inability to use the Model and the Complementary Material (including but not limited to damages for loss of goodwill, work stoppage, computer failure or malfunction, or any and all other commercial damages or losses), even if such Contributor has been advised of the possibility of such damages.
+11. Accepting Warranty or Additional Liability. While redistributing the Model, Derivatives of the Model and the Complementary Material thereof, You may choose to offer, and charge a fee for, acceptance of support, warranty, indemnity, or other liability obligations and/or rights consistent with this License. However, in accepting such obligations, You may act only on Your own behalf and on Your sole responsibility, not on behalf of any other Contributor, and only if You agree to indemnify, defend, and hold each Contributor harmless for any liability incurred by, or claims asserted against, such Contributor by reason of your accepting any such warranty or additional liability.
+12. If any provision of this License is held to be invalid, illegal or unenforceable, the remaining provisions shall be unaffected thereby and remain valid as if such provision had not been set forth herein.
+
+END OF TERMS AND CONDITIONS
+
+
+
+
+Attachment A
+
+Use Restrictions
+
+You agree not to use the Model or Derivatives of the Model:
+- In any way that violates any applicable national, federal, state, local or international law or regulation;
+- For the purpose of exploiting, harming or attempting to exploit or harm minors in any way;
+- To generate or disseminate verifiably false information and/or content with the purpose of harming others;
+- To generate or disseminate personal identifiable information that can be used to harm an individual;
+- To defame, disparage or otherwise harass others;
+- For fully automated decision making that adversely impacts an individual’s legal rights or otherwise creates or modifies a binding, enforceable obligation;
+- For any use intended to or which has the effect of discriminating against or harming individuals or groups based on online or offline social behavior or known or predicted personal or personality characteristics;
+- To exploit any of the vulnerabilities of a specific group of persons based on their age, social, physical or mental characteristics, in order to materially distort the behavior of a person pertaining to that group in a manner that causes or is likely to cause that person or another person physical or psychological harm;
+- For any use intended to or which has the effect of discriminating against individuals or groups based on legally protected characteristics or categories;
+- To provide medical advice and medical results interpretation;
+- To generate or disseminate information for the purpose to be used for administration of justice, law enforcement, immigration or asylum processes, such as predicting an individual will commit fraud/crime commitment (e.g. by text profiling, drawing causal relationships between assertions made in documents, indiscriminate and arbitrarily-targeted use).

Dockerfile

@@ -0,0 +1,15 @@
+FROM pytorch/pytorch:2.1.0-cuda12.1-cudnn8-runtime
+
+ENV TZ=Asia/Shanghai
+
+WORKDIR /app
+
+COPY . /app
+
+RUN pip install --no-cache-dir -r requirements.txt
+
+RUN pip install --no-cache-dir opencv-python-headless -i https://pypi.org/simple
+
+EXPOSE 8888
+
+CMD ["python", "main.py", "--host", "0.0.0.0", "--port", "8888", "--skip-pip"]

LICENSE

@@ -0,0 +1,674 @@
+                    GNU GENERAL PUBLIC LICENSE
+                       Version 3, 29 June 2007
+
+ Copyright (C) 2007 Free Software Foundation, Inc. <https://fsf.org/>
+ Everyone is permitted to copy and distribute verbatim copies
+ of this license document, but changing it is not allowed.
+
+                            Preamble
+
+  The GNU General Public License is a free, copyleft license for
+software and other kinds of works.
+
+  The licenses for most software and other practical works are designed
+to take away your freedom to share and change the works.  By contrast,
+the GNU General Public License is intended to guarantee your freedom to
+share and change all versions of a program--to make sure it remains free
+software for all its users.  We, the Free Software Foundation, use the
+GNU General Public License for most of our software; it applies also to
+any other work released this way by its authors.  You can apply it to
+your programs, too.
+
+  When we speak of free software, we are referring to freedom, not
+price.  Our General Public Licenses are designed to make sure that you
+have the freedom to distribute copies of free software (and charge for
+them if you wish), that you receive source code or can get it if you
+want it, that you can change the software or use pieces of it in new
+free programs, and that you know you can do these things.
+
+  To protect your rights, we need to prevent others from denying you
+these rights or asking you to surrender the rights.  Therefore, you have
+certain responsibilities if you distribute copies of the software, or if
+you modify it: responsibilities to respect the freedom of others.
+
+  For example, if you distribute copies of such a program, whether
+gratis or for a fee, you must pass on to the recipients the same
+freedoms that you received.  You must make sure that they, too, receive
+or can get the source code.  And you must show them these terms so they
+know their rights.
+
+  Developers that use the GNU GPL protect your rights with two steps:
+(1) assert copyright on the software, and (2) offer you this License
+giving you legal permission to copy, distribute and/or modify it.
+
+  For the developers' and authors' protection, the GPL clearly explains
+that there is no warranty for this free software.  For both users' and
+authors' sake, the GPL requires that modified versions be marked as
+changed, so that their problems will not be attributed erroneously to
+authors of previous versions.
+
+  Some devices are designed to deny users access to install or run
+modified versions of the software inside them, although the manufacturer
+can do so.  This is fundamentally incompatible with the aim of
+protecting users' freedom to change the software.  The systematic
+pattern of such abuse occurs in the area of products for individuals to
+use, which is precisely where it is most unacceptable.  Therefore, we
+have designed this version of the GPL to prohibit the practice for those
+products.  If such problems arise substantially in other domains, we
+stand ready to extend this provision to those domains in future versions
+of the GPL, as needed to protect the freedom of users.
+
+  Finally, every program is threatened constantly by software patents.
+States should not allow patents to restrict development and use of
+software on general-purpose computers, but in those that do, we wish to
+avoid the special danger that patents applied to a free program could
+make it effectively proprietary.  To prevent this, the GPL assures that
+patents cannot be used to render the program non-free.
+
+  The precise terms and conditions for copying, distribution and
+modification follow.
+
+                       TERMS AND CONDITIONS
+
+  0. Definitions.
+
+  "This License" refers to version 3 of the GNU General Public License.
+
+  "Copyright" also means copyright-like laws that apply to other kinds of
+works, such as semiconductor masks.
+
+  "The Program" refers to any copyrightable work licensed under this
+License.  Each licensee is addressed as "you".  "Licensees" and
+"recipients" may be individuals or organizations.
+
+  To "modify" a work means to copy from or adapt all or part of the work
+in a fashion requiring copyright permission, other than the making of an
+exact copy.  The resulting work is called a "modified version" of the
+earlier work or a work "based on" the earlier work.
+
+  A "covered work" means either the unmodified Program or a work based
+on the Program.
+
+  To "propagate" a work means to do anything with it that, without
+permission, would make you directly or secondarily liable for
+infringement under applicable copyright law, except executing it on a
+computer or modifying a private copy.  Propagation includes copying,
+distribution (with or without modification), making available to the
+public, and in some countries other activities as well.
+
+  To "convey" a work means any kind of propagation that enables other
+parties to make or receive copies.  Mere interaction with a user through
+a computer network, with no transfer of a copy, is not conveying.
+
+  An interactive user interface displays "Appropriate Legal Notices"
+to the extent that it includes a convenient and prominently visible
+feature that (1) displays an appropriate copyright notice, and (2)
+tells the user that there is no warranty for the work (except to the
+extent that warranties are provided), that licensees may convey the
+work under this License, and how to view a copy of this License.  If
+the interface presents a list of user commands or options, such as a
+menu, a prominent item in the list meets this criterion.
+
+  1. Source Code.
+
+  The "source code" for a work means the preferred form of the work
+for making modifications to it.  "Object code" means any non-source
+form of a work.
+
+  A "Standard Interface" means an interface that either is an official
+standard defined by a recognized standards body, or, in the case of
+interfaces specified for a particular programming language, one that
+is widely used among developers working in that language.
+
+  The "System Libraries" of an executable work include anything, other
+than the work as a whole, that (a) is included in the normal form of
+packaging a Major Component, but which is not part of that Major
+Component, and (b) serves only to enable use of the work with that
+Major Component, or to implement a Standard Interface for which an
+implementation is available to the public in source code form.  A
+"Major Component", in this context, means a major essential component
+(kernel, window system, and so on) of the specific operating system
+(if any) on which the executable work runs, or a compiler used to
+produce the work, or an object code interpreter used to run it.
+
+  The "Corresponding Source" for a work in object code form means all
+the source code needed to generate, install, and (for an executable
+work) run the object code and to modify the work, including scripts to
+control those activities.  However, it does not include the work's
+System Libraries, or general-purpose tools or generally available free
+programs which are used unmodified in performing those activities but
+which are not part of the work.  For example, Corresponding Source
+includes interface definition files associated with source files for
+the work, and the source code for shared libraries and dynamically
+linked subprograms that the work is specifically designed to require,
+such as by intimate data communication or control flow between those
+subprograms and other parts of the work.
+
+  The Corresponding Source need not include anything that users
+can regenerate automatically from other parts of the Corresponding
+Source.
+
+  The Corresponding Source for a work in source code form is that
+same work.
+
+  2. Basic Permissions.
+
+  All rights granted under this License are granted for the term of
+copyright on the Program, and are irrevocable provided the stated
+conditions are met.  This License explicitly affirms your unlimited
+permission to run the unmodified Program.  The output from running a
+covered work is covered by this License only if the output, given its
+content, constitutes a covered work.  This License acknowledges your
+rights of fair use or other equivalent, as provided by copyright law.
+
+  You may make, run and propagate covered works that you do not
+convey, without conditions so long as your license otherwise remains
+in force.  You may convey covered works to others for the sole purpose
+of having them make modifications exclusively for you, or provide you
+with facilities for running those works, provided that you comply with
+the terms of this License in conveying all material for which you do
+not control copyright.  Those thus making or running the covered works
+for you must do so exclusively on your behalf, under your direction
+and control, on terms that prohibit them from making any copies of
+your copyrighted material outside their relationship with you.
+
+  Conveying under any other circumstances is permitted solely under
+the conditions stated below.  Sublicensing is not allowed; section 10
+makes it unnecessary.
+
+  3. Protecting Users' Legal Rights From Anti-Circumvention Law.
+
+  No covered work shall be deemed part of an effective technological
+measure under any applicable law fulfilling obligations under article
+11 of the WIPO copyright treaty adopted on 20 December 1996, or
+similar laws prohibiting or restricting circumvention of such
+measures.
+
+  When you convey a covered work, you waive any legal power to forbid
+circumvention of technological measures to the extent such circumvention
+is effected by exercising rights under this License with respect to
+the covered work, and you disclaim any intention to limit operation or
+modification of the work as a means of enforcing, against the work's
+users, your or third parties' legal rights to forbid circumvention of
+technological measures.
+
+  4. Conveying Verbatim Copies.
+
+  You may convey verbatim copies of the Program's source code as you
+receive it, in any medium, provided that you conspicuously and
+appropriately publish on each copy an appropriate copyright notice;
+keep intact all notices stating that this License and any
+non-permissive terms added in accord with section 7 apply to the code;
+keep intact all notices of the absence of any warranty; and give all
+recipients a copy of this License along with the Program.
+
+  You may charge any price or no price for each copy that you convey,
+and you may offer support or warranty protection for a fee.
+
+  5. Conveying Modified Source Versions.
+
+  You may convey a work based on the Program, or the modifications to
+produce it from the Program, in the form of source code under the
+terms of section 4, provided that you also meet all of these conditions:
+
+    a) The work must carry prominent notices stating that you modified
+    it, and giving a relevant date.
+
+    b) The work must carry prominent notices stating that it is
+    released under this License and any conditions added under section
+    7.  This requirement modifies the requirement in section 4 to
+    "keep intact all notices".
+
+    c) You must license the entire work, as a whole, under this
+    License to anyone who comes into possession of a copy.  This
+    License will therefore apply, along with any applicable section 7
+    additional terms, to the whole of the work, and all its parts,
+    regardless of how they are packaged.  This License gives no
+    permission to license the work in any other way, but it does not
+    invalidate such permission if you have separately received it.
+
+    d) If the work has interactive user interfaces, each must display
+    Appropriate Legal Notices; however, if the Program has interactive
+    interfaces that do not display Appropriate Legal Notices, your
+    work need not make them do so.
+
+  A compilation of a covered work with other separate and independent
+works, which are not by their nature extensions of the covered work,
+and which are not combined with it such as to form a larger program,
+in or on a volume of a storage or distribution medium, is called an
+"aggregate" if the compilation and its resulting copyright are not
+used to limit the access or legal rights of the compilation's users
+beyond what the individual works permit.  Inclusion of a covered work
+in an aggregate does not cause this License to apply to the other
+parts of the aggregate.
+
+  6. Conveying Non-Source Forms.
+
+  You may convey a covered work in object code form under the terms
+of sections 4 and 5, provided that you also convey the
+machine-readable Corresponding Source under the terms of this License,
+in one of these ways:
+
+    a) Convey the object code in, or embodied in, a physical product
+    (including a physical distribution medium), accompanied by the
+    Corresponding Source fixed on a durable physical medium
+    customarily used for software interchange.
+
+    b) Convey the object code in, or embodied in, a physical product
+    (including a physical distribution medium), accompanied by a
+    written offer, valid for at least three years and valid for as
+    long as you offer spare parts or customer support for that product
+    model, to give anyone who possesses the object code either (1) a
+    copy of the Corresponding Source for all the software in the
+    product that is covered by this License, on a durable physical
+    medium customarily used for software interchange, for a price no
+    more than your reasonable cost of physically performing this
+    conveying of source, or (2) access to copy the
+    Corresponding Source from a network server at no charge.
+
+    c) Convey individual copies of the object code with a copy of the
+    written offer to provide the Corresponding Source.  This
+    alternative is allowed only occasionally and noncommercially, and
+    only if you received the object code with such an offer, in accord
+    with subsection 6b.
+
+    d) Convey the object code by offering access from a designated
+    place (gratis or for a charge), and offer equivalent access to the
+    Corresponding Source in the same way through the same place at no
+    further charge.  You need not require recipients to copy the
+    Corresponding Source along with the object code.  If the place to
+    copy the object code is a network server, the Corresponding Source
+    may be on a different server (operated by you or a third party)
+    that supports equivalent copying facilities, provided you maintain
+    clear directions next to the object code saying where to find the
+    Corresponding Source.  Regardless of what server hosts the
+    Corresponding Source, you remain obligated to ensure that it is
+    available for as long as needed to satisfy these requirements.
+
+    e) Convey the object code using peer-to-peer transmission, provided
+    you inform other peers where the object code and Corresponding
+    Source of the work are being offered to the general public at no
+    charge under subsection 6d.
+
+  A separable portion of the object code, whose source code is excluded
+from the Corresponding Source as a System Library, need not be
+included in conveying the object code work.
+
+  A "User Product" is either (1) a "consumer product", which means any
+tangible personal property which is normally used for personal, family,
+or household purposes, or (2) anything designed or sold for incorporation
+into a dwelling.  In determining whether a product is a consumer product,
+doubtful cases shall be resolved in favor of coverage.  For a particular
+product received by a particular user, "normally used" refers to a
+typical or common use of that class of product, regardless of the status
+of the particular user or of the way in which the particular user
+actually uses, or expects or is expected to use, the product.  A product
+is a consumer product regardless of whether the product has substantial
+commercial, industrial or non-consumer uses, unless such uses represent
+the only significant mode of use of the product.
+
+  "Installation Information" for a User Product means any methods,
+procedures, authorization keys, or other information required to install
+and execute modified versions of a covered work in that User Product from
+a modified version of its Corresponding Source.  The information must
+suffice to ensure that the continued functioning of the modified object
+code is in no case prevented or interfered with solely because
+modification has been made.
+
+  If you convey an object code work under this section in, or with, or
+specifically for use in, a User Product, and the conveying occurs as
+part of a transaction in which the right of possession and use of the
+User Product is transferred to the recipient in perpetuity or for a
+fixed term (regardless of how the transaction is characterized), the
+Corresponding Source conveyed under this section must be accompanied
+by the Installation Information.  But this requirement does not apply
+if neither you nor any third party retains the ability to install
+modified object code on the User Product (for example, the work has
+been installed in ROM).
+
+  The requirement to provide Installation Information does not include a
+requirement to continue to provide support service, warranty, or updates
+for a work that has been modified or installed by the recipient, or for
+the User Product in which it has been modified or installed.  Access to a
+network may be denied when the modification itself materially and
+adversely affects the operation of the network or violates the rules and
+protocols for communication across the network.
+
+  Corresponding Source conveyed, and Installation Information provided,
+in accord with this section must be in a format that is publicly
+documented (and with an implementation available to the public in
+source code form), and must require no special password or key for
+unpacking, reading or copying.
+
+  7. Additional Terms.
+
+  "Additional permissions" are terms that supplement the terms of this
+License by making exceptions from one or more of its conditions.
+Additional permissions that are applicable to the entire Program shall
+be treated as though they were included in this License, to the extent
+that they are valid under applicable law.  If additional permissions
+apply only to part of the Program, that part may be used separately
+under those permissions, but the entire Program remains governed by
+this License without regard to the additional permissions.
+
+  When you convey a copy of a covered work, you may at your option
+remove any additional permissions from that copy, or from any part of
+it.  (Additional permissions may be written to require their own
+removal in certain cases when you modify the work.)  You may place
+additional permissions on material, added by you to a covered work,
+for which you have or can give appropriate copyright permission.
+
+  Notwithstanding any other provision of this License, for material you
+add to a covered work, you may (if authorized by the copyright holders of
+that material) supplement the terms of this License with terms:
+
+    a) Disclaiming warranty or limiting liability differently from the
+    terms of sections 15 and 16 of this License; or
+
+    b) Requiring preservation of specified reasonable legal notices or
+    author attributions in that material or in the Appropriate Legal
+    Notices displayed by works containing it; or
+
+    c) Prohibiting misrepresentation of the origin of that material, or
+    requiring that modified versions of such material be marked in
+    reasonable ways as different from the original version; or
+
+    d) Limiting the use for publicity purposes of names of licensors or
+    authors of the material; or
+
+    e) Declining to grant rights under trademark law for use of some
+    trade names, trademarks, or service marks; or
+
+    f) Requiring indemnification of licensors and authors of that
+    material by anyone who conveys the material (or modified versions of
+    it) with contractual assumptions of liability to the recipient, for
+    any liability that these contractual assumptions directly impose on
+    those licensors and authors.
+
+  All other non-permissive additional terms are considered "further
+restrictions" within the meaning of section 10.  If the Program as you
+received it, or any part of it, contains a notice stating that it is
+governed by this License along with a term that is a further
+restriction, you may remove that term.  If a license document contains
+a further restriction but permits relicensing or conveying under this
+License, you may add to a covered work material governed by the terms
+of that license document, provided that the further restriction does
+not survive such relicensing or conveying.
+
+  If you add terms to a covered work in accord with this section, you
+must place, in the relevant source files, a statement of the
+additional terms that apply to those files, or a notice indicating
+where to find the applicable terms.
+
+  Additional terms, permissive or non-permissive, may be stated in the
+form of a separately written license, or stated as exceptions;
+the above requirements apply either way.
+
+  8. Termination.
+
+  You may not propagate or modify a covered work except as expressly
+provided under this License.  Any attempt otherwise to propagate or
+modify it is void, and will automatically terminate your rights under
+this License (including any patent licenses granted under the third
+paragraph of section 11).
+
+  However, if you cease all violation of this License, then your
+license from a particular copyright holder is reinstated (a)
+provisionally, unless and until the copyright holder explicitly and
+finally terminates your license, and (b) permanently, if the copyright
+holder fails to notify you of the violation by some reasonable means
+prior to 60 days after the cessation.
+
+  Moreover, your license from a particular copyright holder is
+reinstated permanently if the copyright holder notifies you of the
+violation by some reasonable means, this is the first time you have
+received notice of violation of this License (for any work) from that
+copyright holder, and you cure the violation prior to 30 days after
+your receipt of the notice.
+
+  Termination of your rights under this section does not terminate the
+licenses of parties who have received copies or rights from you under
+this License.  If your rights have been terminated and not permanently
+reinstated, you do not qualify to receive new licenses for the same
+material under section 10.
+
+  9. Acceptance Not Required for Having Copies.
+
+  You are not required to accept this License in order to receive or
+run a copy of the Program.  Ancillary propagation of a covered work
+occurring solely as a consequence of using peer-to-peer transmission
+to receive a copy likewise does not require acceptance.  However,
+nothing other than this License grants you permission to propagate or
+modify any covered work.  These actions infringe copyright if you do
+not accept this License.  Therefore, by modifying or propagating a
+covered work, you indicate your acceptance of this License to do so.
+
+  10. Automatic Licensing of Downstream Recipients.
+
+  Each time you convey a covered work, the recipient automatically
+receives a license from the original licensors, to run, modify and
+propagate that work, subject to this License.  You are not responsible
+for enforcing compliance by third parties with this License.
+
+  An "entity transaction" is a transaction transferring control of an
+organization, or substantially all assets of one, or subdividing an
+organization, or merging organizations.  If propagation of a covered
+work results from an entity transaction, each party to that
+transaction who receives a copy of the work also receives whatever
+licenses to the work the party's predecessor in interest had or could
+give under the previous paragraph, plus a right to possession of the
+Corresponding Source of the work from the predecessor in interest, if
+the predecessor has it or can get it with reasonable efforts.
+
+  You may not impose any further restrictions on the exercise of the
+rights granted or affirmed under this License.  For example, you may
+not impose a license fee, royalty, or other charge for exercise of
+rights granted under this License, and you may not initiate litigation
+(including a cross-claim or counterclaim in a lawsuit) alleging that
+any patent claim is infringed by making, using, selling, offering for
+sale, or importing the Program or any portion of it.
+
+  11. Patents.
+
+  A "contributor" is a copyright holder who authorizes use under this
+License of the Program or a work on which the Program is based.  The
+work thus licensed is called the contributor's "contributor version".
+
+  A contributor's "essential patent claims" are all patent claims
+owned or controlled by the contributor, whether already acquired or
+hereafter acquired, that would be infringed by some manner, permitted
+by this License, of making, using, or selling its contributor version,
+but do not include claims that would be infringed only as a
+consequence of further modification of the contributor version.  For
+purposes of this definition, "control" includes the right to grant
+patent sublicenses in a manner consistent with the requirements of
+this License.
+
+  Each contributor grants you a non-exclusive, worldwide, royalty-free
+patent license under the contributor's essential patent claims, to
+make, use, sell, offer for sale, import and otherwise run, modify and
+propagate the contents of its contributor version.
+
+  In the following three paragraphs, a "patent license" is any express
+agreement or commitment, however denominated, not to enforce a patent
+(such as an express permission to practice a patent or covenant not to
+sue for patent infringement).  To "grant" such a patent license to a
+party means to make such an agreement or commitment not to enforce a
+patent against the party.
+
+  If you convey a covered work, knowingly relying on a patent license,
+and the Corresponding Source of the work is not available for anyone
+to copy, free of charge and under the terms of this License, through a
+publicly available network server or other readily accessible means,
+then you must either (1) cause the Corresponding Source to be so
+available, or (2) arrange to deprive yourself of the benefit of the
+patent license for this particular work, or (3) arrange, in a manner
+consistent with the requirements of this License, to extend the patent
+license to downstream recipients.  "Knowingly relying" means you have
+actual knowledge that, but for the patent license, your conveying the
+covered work in a country, or your recipient's use of the covered work
+in a country, would infringe one or more identifiable patents in that
+country that you have reason to believe are valid.
+
+  If, pursuant to or in connection with a single transaction or
+arrangement, you convey, or propagate by procuring conveyance of, a
+covered work, and grant a patent license to some of the parties
+receiving the covered work authorizing them to use, propagate, modify
+or convey a specific copy of the covered work, then the patent license
+you grant is automatically extended to all recipients of the covered
+work and works based on it.
+
+  A patent license is "discriminatory" if it does not include within
+the scope of its coverage, prohibits the exercise of, or is
+conditioned on the non-exercise of one or more of the rights that are
+specifically granted under this License.  You may not convey a covered
+work if you are a party to an arrangement with a third party that is
+in the business of distributing software, under which you make payment
+to the third party based on the extent of your activity of conveying
+the work, and under which the third party grants, to any of the
+parties who would receive the covered work from you, a discriminatory
+patent license (a) in connection with copies of the covered work
+conveyed by you (or copies made from those copies), or (b) primarily
+for and in connection with specific products or compilations that
+contain the covered work, unless you entered into that arrangement,
+or that patent license was granted, prior to 28 March 2007.
+
+  Nothing in this License shall be construed as excluding or limiting
+any implied license or other defenses to infringement that may
+otherwise be available to you under applicable patent law.
+
+  12. No Surrender of Others' Freedom.
+
+  If conditions are imposed on you (whether by court order, agreement or
+otherwise) that contradict the conditions of this License, they do not
+excuse you from the conditions of this License.  If you cannot convey a
+covered work so as to satisfy simultaneously your obligations under this
+License and any other pertinent obligations, then as a consequence you may
+not convey it at all.  For example, if you agree to terms that obligate you
+to collect a royalty for further conveying from those to whom you convey
+the Program, the only way you could satisfy both those terms and this
+License would be to refrain entirely from conveying the Program.
+
+  13. Use with the GNU Affero General Public License.
+
+  Notwithstanding any other provision of this License, you have
+permission to link or combine any covered work with a work licensed
+under version 3 of the GNU Affero General Public License into a single
+combined work, and to convey the resulting work.  The terms of this
+License will continue to apply to the part which is the covered work,
+but the special requirements of the GNU Affero General Public License,
+section 13, concerning interaction through a network will apply to the
+combination as such.
+
+  14. Revised Versions of this License.
+
+  The Free Software Foundation may publish revised and/or new versions of
+the GNU General Public License from time to time.  Such new versions will
+be similar in spirit to the present version, but may differ in detail to
+address new problems or concerns.
+
+  Each version is given a distinguishing version number.  If the
+Program specifies that a certain numbered version of the GNU General
+Public License "or any later version" applies to it, you have the
+option of following the terms and conditions either of that numbered
+version or of any later version published by the Free Software
+Foundation.  If the Program does not specify a version number of the
+GNU General Public License, you may choose any version ever published
+by the Free Software Foundation.
+
+  If the Program specifies that a proxy can decide which future
+versions of the GNU General Public License can be used, that proxy's
+public statement of acceptance of a version permanently authorizes you
+to choose that version for the Program.
+
+  Later license versions may give you additional or different
+permissions.  However, no additional obligations are imposed on any
+author or copyright holder as a result of your choosing to follow a
+later version.
+
+  15. Disclaimer of Warranty.
+
+  THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY
+APPLICABLE LAW.  EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT
+HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY
+OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO,
+THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+PURPOSE.  THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM
+IS WITH YOU.  SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF
+ALL NECESSARY SERVICING, REPAIR OR CORRECTION.
+
+  16. Limitation of Liability.
+
+  IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING
+WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES AND/OR CONVEYS
+THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, INCLUDING ANY
+GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE
+USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF
+DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD
+PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER PROGRAMS),
+EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF
+SUCH DAMAGES.
+
+  17. Interpretation of Sections 15 and 16.
+
+  If the disclaimer of warranty and limitation of liability provided
+above cannot be given local legal effect according to their terms,
+reviewing courts shall apply local law that most closely approximates
+an absolute waiver of all civil liability in connection with the
+Program, unless a warranty or assumption of liability accompanies a
+copy of the Program in return for a fee.
+
+                     END OF TERMS AND CONDITIONS
+
+            How to Apply These Terms to Your New Programs
+
+  If you develop a new program, and you want it to be of the greatest
+possible use to the public, the best way to achieve this is to make it
+free software which everyone can redistribute and change under these terms.
+
+  To do so, attach the following notices to the program.  It is safest
+to attach them to the start of each source file to most effectively
+state the exclusion of warranty; and each file should have at least
+the "copyright" line and a pointer to where the full notice is found.
+
+    <one line to give the program's name and a brief idea of what it does.>
+    Copyright (C) <year>  <name of author>
+
+    This program is free software: you can redistribute it and/or modify
+    it under the terms of the GNU General Public License as published by
+    the Free Software Foundation, either version 3 of the License, or
+    (at your option) any later version.
+
+    This program is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    GNU General Public License for more details.
+
+    You should have received a copy of the GNU General Public License
+    along with this program.  If not, see <https://www.gnu.org/licenses/>.
+
+Also add information on how to contact you by electronic and paper mail.
+
+  If the program does terminal interaction, make it output a short
+notice like this when it starts in an interactive mode:
+
+    <program>  Copyright (C) <year>  <name of author>
+    This program comes with ABSOLUTELY NO WARRANTY; for details type `show w'.
+    This is free software, and you are welcome to redistribute it
+    under certain conditions; type `show c' for details.
+
+The hypothetical commands `show w' and `show c' should show the appropriate
+parts of the General Public License.  Of course, your program's commands
+might be different; for a GUI interface, you would use an "about box".
+
+  You should also get your employer (if you work as a programmer) or school,
+if any, to sign a "copyright disclaimer" for the program, if necessary.
+For more information on this, and how to apply and follow the GNU GPL, see
+<https://www.gnu.org/licenses/>.
+
+  The GNU General Public License does not permit incorporating your program
+into proprietary programs.  If your program is a subroutine library, you
+may consider it more useful to permit linking proprietary applications with
+the library.  If this is what you want to do, use the GNU Lesser General
+Public License instead of this License.  But first, please read
+<https://www.gnu.org/licenses/why-not-lgpl.html>.

README copy.md

@@ -0,0 +1,286 @@
+[![Docker Image CI](https://github.com/konieshadow/Fooocus-API/actions/workflows/docker-image.yml/badge.svg?branch=main)](https://github.com/konieshadow/Fooocus-API/actions/workflows/docker-image.yml)
+
+[ English | [中文](/README_zh.md) ]
+
+- [Introduction](#introduction)
+  - [Fooocus](#fooocus)
+  - [Fooocus-API](#fooocus-api)
+- [Get-Start](#get-start)
+  - [Run with Replicate](#run-with-replicate)
+  - [Self-hosted](#self-hosted)
+    - [conda](#conda)
+    - [venv](#venv)
+    - [predownload and install](#predownload-and-install)
+    - [already exist Fooocus](#already-exist-fooocus)
+  - [Start with docker](#start-with-docker)
+- [cmd flags](#cmd-flags)
+- [Change log](#change-log)
+- [Apis](#apis)
+- [License](#license)
+- [Thanks :purple\_heart:](#thanks-purple_heart)
+
+> Note:
+>
+> Although I tested it, I still suggest you test it again before the official update
+>
+> Fooocus 2.5 includes a significant update, with most dependencies upgraded. Therefore, after updating, do not use `--skip-pip` unless you have already performed a manual update.
+>
+> Additionally, `groundingdino-py` may encounter installation errors, especially in Chinese Windows environments. The solution can be found in the following [issue](https://github.com/IDEA-Research/GroundingDINO/issues/206).
+
+
+> GenerateMask is same as DescribeImage, It is not process as a task, result will directly return
+
+# Instructions for Using the ImageEnhance Interface
+Below are examples of parameters that include the main parameters required for ImageEnhance. The V1 interface adopts a form-like approach similar to ImagePrompt to break down the enhance controller.
+
+
+```python
+{
+  "enhance_input_image": "",
+  "enhance_checkbox": true,
+  "enhance_uov_method": "Vary (Strong)",
+  "enhance_uov_processing_order": "Before First Enhancement",
+  "enhance_uov_prompt_type": "Original Prompts",
+  "save_final_enhanced_image_only": true,
+  "enhance_ctrlnets": [
+    {
+      "enhance_enabled": false,
+      "enhance_mask_dino_prompt": "face",
+      "enhance_prompt": "",
+      "enhance_negative_prompt": "",
+      "enhance_mask_model": "sam",
+      "enhance_mask_cloth_category": "full",
+      "enhance_mask_sam_model": "vit_b",
+      "enhance_mask_text_threshold": 0.25,
+      "enhance_mask_box_threshold": 0.3,
+      "enhance_mask_sam_max_detections": 0,
+      "enhance_inpaint_disable_initial_latent": false,
+      "enhance_inpaint_engine": "v2.6",
+      "enhance_inpaint_strength": 1,
+      "enhance_inpaint_respective_field": 0.618,
+      "enhance_inpaint_erode_or_dilate": 0,
+      "enhance_mask_invert": false
+    }
+  ]
+}
+```
+
+- enhance_input_image: The image to be enhanced, which is required and can be provided as an image URL for the V2 interface.
+- enhance_checkbox: A toggle switch that must be set to true if you want to use the enhance image feature.
+- save_final_enhanced_image_only: Since image enhancement is a pipeline operation, it can produce multiple result images. This parameter allows you to only return the final enhanced image.
+
+There are three parameters related to UpscaleVary, which are used to perform Upscale or Vary before or after enhancement.
+
+- enhance_uov_method: Similar to the UpscaleOrVary interface, Disabled turns it off.
+- enhance_uov_processing_order: Determines whether to process the image before or after enhancement.
+- enhance_uov_prompt_type: I'm not sure about the specific function; you might want to research it based on the WebUI.
+
+The `enhance_ctrlnets` element is a list of ImageEnhance controller objects, with a maximum of three elements in the list, any additional elements will be discarded. The parameters correspond roughly to the WebUI, and the notable parameters are:
+
+- enhance_enabled: This parameter controls whether the enhance controller is active. If there are no enabled enhance controllers, the task will be skipped.
+- enhance_mask_dino_prompt: This parameter is required and indicates the area to be enhanced. If it is empty, even if the enhance controller is enabled, the task will be skipped.
+
+
+# Introduction
+
+FastAPI powered API for [Fooocus](https://github.com/lllyasviel/Fooocus).
+
+Currently loaded Fooocus version: [2.3.0](https://github.com/lllyasviel/Fooocus/blob/main/update_log.md).
+
+## Fooocus
+
+This part from [Fooocus](https://github.com/lllyasviel/Fooocus) project.
+
+Fooocus is an image generating software (based on [Gradio](https://www.gradio.app/)).
+
+Fooocus is a rethinking of Stable Diffusion and Midjourney’s designs:
+
+- Learned from Stable Diffusion, the software is offline, open source, and free.
+
+- Learned from Midjourney, the manual tweaking is not needed, and users only need to focus on the prompts and images.
+
+Fooocus has included and automated lots of inner optimizations and quality improvements. Users can forget all those difficult technical parameters, and just enjoy the interaction between human and computer to "explore new mediums of thought and expanding the imaginative powers of the human species"
+
+## Fooocus-API
+
+I think you must have tried to use [Gradio client](https://www.gradio.app/docs/client) to call Fooocus, which was a terrible experience for me. 
+
+Fooocus API uses [FastAPI](https://fastapi.tiangolo.com/)  provides the `REST` API for using Fooocus. Now, you can use Fooocus's powerful ability in any language you like. 
+
+In addition, we also provide detailed [documentation](/docs/api_doc_en.md) and [sample code](/examples)
+
+# Get-Start
+
+## Run with Replicate
+
+Now you can use Fooocus-API by Replicate, the model is on [konieshadow/fooocus-api](https://replicate.com/konieshadow/fooocus-api).
+
+With preset:
+
+- [konieshadow/fooocus-api-anime](https://replicate.com/konieshadow/fooocus-api-anime)
+- [konieshadow/fooocus-api-realistic](https://replicate.com/konieshadow/fooocus-api-realistic)
+
+I believe this is the easiest way to generate image with Fooocus's power.
+
+## Self-hosted
+
+You need python version >= 3.10, or use conda to create a new env.
+
+The hardware requirements are what Fooocus needs. You can find detail [here](https://github.com/lllyasviel/Fooocus#minimal-requirement)
+
+### conda
+
+You can easily start app follow this step use conda:
+
+```shell
+conda env create -f environment.yaml
+conda activate fooocus-api
+```
+
+and then, run `python main.py` to start app, default, server is listening on `http://127.0.0.1:8888`
+
+> If you are running the project for the first time, you may have to wait for a while, during which time the program will complete the rest of the installation and download the necessary models. You can also do these steps manually, which I'll mention later.
+
+### venv
+
+Similar to using conda, create a virtual environment, and then start and wait for a while
+
+```powershell
+# windows
+python -m venv venv
+.\venv\Scripts\Activate
+```
+
+```shell
+# linux
+python -m venv venv
+source venv/bin/activate
+```
+and then, run `python main.py`
+
+### predownload and install
+
+If you want to deal with environmental problems manually and download the model in advance, you can refer to the following steps
+
+After creating a complete environment using conda or venv, you can manually complete the installation of the subsequent environment, just follow
+
+first, install requirements: `pip install -r requirements.txt`
+
+then, pytorch with cuda: `pip install torch==2.1.0 torchvision==0.16.0 torchaudio==2.1.0 --index-url https://download.pytorch.org/whl/cu121` , you can find more info about this [here](https://pytorch.org/get-started/previous-versions/),
+
+> It is important to note that for pytorch and cuda versions, the recommended version of Fooocus is used, which is currently pytorch2.1.0+cuda12.1. If you insist, you can also use other versions, but you need to add `--skip-pip` when you start app, otherwise the recommended version will be installed automatically
+
+Go to the `repositories` directories, download models and put it into `repositories\Fooocus\models`
+
+If you have Fooocus installed, see [already-exist-fooocus](#already-exist-fooocus)
+
+here is a list need to download for startup (for different [startup params](#cmd-flags) maybe difference):
+
+- checkpoint:  path to `repositories\Fooocus\models\checkpoints`
+    + [juggernautXL_version6Rundiffusion.safetensors](https://huggingface.co/lllyasviel/fav_models/resolve/main/fav/juggernautXL_version6Rundiffusion.safetensors)
+
+- vae_approx: path to `repositories\Fooocus\models\vae_approx`
+    + [xlvaeapp.pth](https://huggingface.co/lllyasviel/misc/resolve/main/xlvaeapp.pth)
+    + [vaeapp_sd15.pth](https://huggingface.co/lllyasviel/misc/resolve/main/vaeapp_sd15.pt)
+    + [xl-to-v1_interposer-v3.1.safetensors](https://huggingface.co/lllyasviel/misc/resolve/main/xl-to-v1_interposer-v3.1.safetensors)
+
+- lora: path to `repositories\Fooocus\models\loras`
+    + [sd_xl_offset_example-lora_1.0.safetensors](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0/resolve/main/sd_xl_offset_example-lora_1.0.safetensors?download=true)
+
+> I've uploaded the model I'm using, which contains almost all the base models that Fooocus will use! I put it [here](https://www.123pan.com/s/dF5A-SIQsh.html) 提取码: `D4Mk`
+
+### already exist Fooocus
+
+If you already have Fooocus installed, and it is work well, The recommended way is to reuse models, you just simple copy `config.txt` file from your local Fooocus folder to Fooocus-API root folder. See [Customization](https://github.com/lllyasviel/Fooocus#customization) for details.
+
+Use this method you will have both Fooocus and Fooocus-API running at the same time. And they operate independently and do not interfere with each other.
+
+> Do not copy Fooocus to repositories directory
+
+## Start with docker
+
+Before use docker with GPU, you should [install NVIDIA Container Toolkit](https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/latest/install-guide.html) first.
+
+Run
+
+```shell
+docker run -d --gpus=all \
+    -e NVIDIA_DRIVER_CAPABILITIES=compute,utility \
+    -e NVIDIA_VISIBLE_DEVICES=all \
+    -p 8888:8888 konieshadow/fooocus-api
+```
+
+For a more complex usage:
+
+```shell
+mkdir ~/repositories
+mkdir -p ~/.cache/pip
+
+docker run -d --gpus=all \
+    -e NVIDIA_DRIVER_CAPABILITIES=compute,utility \
+    -e NVIDIA_VISIBLE_DEVICES=all \
+    -v ~/repositories:/app/repositories \
+    -v ~/.cache/pip:/root/.cache/pip \
+    -p 8888:8888 konieshadow/fooocus-api
+```
+
+It will be persistent the dependent repositories and pip cache.
+
+You can add `-e PIP_INDEX_URL={pypi-mirror-url}` to docker run command to change pip index url.
+
+> From version 0.4.0.0, Full environment include in docker image, mapping `models` or project root if you needed
+> For example:
+> ```
+> docker run -d --gpus all \
+>     -v /Fooocus-API:/app \
+>     -p 8888:8888 konieshadow/fooocus-api
+>```
+
+# cmd flags
+
+- `-h, --help` show this help message and exit
+- `--port PORT` Set the listen port, default: 8888
+- `--host HOST` Set the listen host, default: 127.0.0.1
+- `--base-url BASE_URL` Set base url for outside visit, default is http://host:port
+- `--log-level LOG_LEVEL` Log info for Uvicorn, default: info
+- `--skip-pip` Skip automatic pip install when setup
+- `--preload-pipeline` Preload pipeline before start http server
+- `--queue-size QUEUE_SIZE` Working queue size, default: 100, generation requests exceeding working queue size will return failure
+- `--queue-history QUEUE_HISTORY` Finished jobs reserve size, tasks exceeding the limit will be deleted, including output image files, default: 0, means no limit
+- `--webhook-url WEBHOOK_URL` Webhook url for notify generation result, default: None
+- `--persistent` Store history to db
+- `--apikey APIKEY` Set apikey to enable secure api, default: None
+
+Since v0.3.25, added CMD flags support of Fooocus. You can pass any argument which Fooocus supported.
+
+For example, to startup image generation (need more vRAM):
+
+```
+python main.py --all-in-fp16 --always-gpu
+```
+
+For Fooocus CMD flags, see [here](https://github.com/lllyasviel/Fooocus?tab=readme-ov-file#all-cmd-flags).
+
+
+# Change log
+
+[CHANGELOG](./docs/change_logs.md)
+
+older change history you can find in [release page](https://github.com/konieshadow/Fooocus-API/releases)
+
+
+# Apis
+
+you can find all api detail [here](/docs/api_doc_en.md)
+
+# License
+
+This repository is licensed under the [GUN General Public License v3.0](https://github.com/mrhan1993/Fooocus-API/blob/main/LICENSE)
+
+The default checkpoint is published by [RunDiffusion](https://huggingface.co/RunDiffusion), is licensed under the [CreativeML Open RAIL-M](https://github.com/mrhan1993/Fooocus-API/blob/main/CreativeMLOpenRAIL-M).
+
+or, you can find it [here](https://huggingface.co/spaces/CompVis/stable-diffusion-license)
+
+# Thanks :purple_heart:
+
+Thanks for all your contributions and efforts towards improving the Fooocus API. We thank you for being part of our :sparkles: community :sparkles:!

README_zh.md

@@ -0,0 +1,288 @@
+[![Docker Image CI](https://github.com/konieshadow/Fooocus-API/actions/workflows/docker-image.yml/badge.svg?branch=main)](https://github.com/konieshadow/Fooocus-API/actions/workflows/docker-image.yml)
+
+[ [English](/README.md) | 中文 ]
+
+- [简介](#简介)
+  - [Fooocus](#fooocus)
+  - [Fooocus-API](#fooocus-api)
+- [开始](#开始)
+  - [在 Replicate 上运行](#在-replicate-上运行)
+  - [自托管](#自托管)
+    - [conda](#conda)
+    - [venv](#venv)
+    - [预下载及安装](#预下载及安装)
+    - [已经有安装好的 Fooocus](#已经有安装好的-fooocus)
+  - [使用Docker启动](#使用docker启动)
+- [命令行参数](#命令行参数)
+- [更新日志](#更新日志)
+- [Apis](#apis)
+- [License](#license)
+- [感谢 :purple\_heart:](#感谢-purple_heart)
+
+
+> 注意：
+>
+> 尽管我进行了测试，但我仍建议你在正式更新前再测一遍
+>
+> Fooocus 2.5 包含大量更新，其中多数依赖进行了升级，因此，更新后请不要使用 `--skip-pip`. 除非你已经进行过手动更新
+>
+> 此外, `groundingdino-py` 可能会遇到安装错误, 特别是在中文 windows 环境中, 解决办法参考: [issues](https://github.com/IDEA-Research/GroundingDINO/issues/206)
+
+> 和 DescribeImage 一样，GenerateMask 不会作为 task 处理而是直接返回结果
+
+# ImageEnhance 接口的使用说明
+
+以下面的参数为例，它包含了 ImageEnhance 所需要的主要参数，V1 接口采用和 ImagePrompt 类似的方式将 enhance 控制器拆分成表单形式：
+
+```python
+{
+  "enhance_input_image": "",
+  "enhance_checkbox": true,
+  "enhance_uov_method": "Vary (Strong)",
+  "enhance_uov_processing_order": "Before First Enhancement",
+  "enhance_uov_prompt_type": "Original Prompts",
+  "save_final_enhanced_image_only": true,
+  "enhance_ctrlnets": [
+    {
+      "enhance_enabled": false,
+      "enhance_mask_dino_prompt": "face",
+      "enhance_prompt": "",
+      "enhance_negative_prompt": "",
+      "enhance_mask_model": "sam",
+      "enhance_mask_cloth_category": "full",
+      "enhance_mask_sam_model": "vit_b",
+      "enhance_mask_text_threshold": 0.25,
+      "enhance_mask_box_threshold": 0.3,
+      "enhance_mask_sam_max_detections": 0,
+      "enhance_inpaint_disable_initial_latent": false,
+      "enhance_inpaint_engine": "v2.6",
+      "enhance_inpaint_strength": 1,
+      "enhance_inpaint_respective_field": 0.618,
+      "enhance_inpaint_erode_or_dilate": 0,
+      "enhance_mask_invert": false
+    }
+  ]
+}
+```
+
+- enhance_input_image：需要增强的图像，如果是 v2 接口，可以提供一个图像 url，必选
+- enhance_checkbox：总开关，使用 enhance image 必须设置为 true
+- save_final_enhanced_image_only：图像增强是一个管道作业，因此会产生多个结果图像，使用该参数仅返回最终图像
+
+有三个和 UpscaleVary 相关的参数，其作用是执行增强之前或完成增强之后执行 Upscale 或 Vary
+
+- enhance_uov_method：和 UpscaleOrVary 接口一样，Disabled 是关闭
+- enhance_uov_processing_order：在增强之前处理还是处理增强后的图像
+- enhance_uov_prompt_type：我也不知道具体作用，对着 WebUI 研究研究🧐
+
+`enhance_ctrlnets` 元素为 ImageEnhance 控制器对象列表，该列表最多包含 3 个元素，多余会被丢弃。参数和 WebUI 基本一一对应，需要注意的参数是：
+
+- enhance_enabled：参数控制该 enhance 控制器是否工作，如果没有开启的 enhance 控制器，任务会被跳过
+- enhance_mask_dino_prompt：该参数必选，表示需要增强的部位，如果该参数为空，即便 enhance 控制器处于开启状态，也会跳过
+
+# 简介
+
+使用 FastAPI 构建的 [Fooocus](https://github.com/lllyasviel/Fooocus) 的 API。
+
+当前支持的 Fooocus 版本: [2.5.3](https://github.com/lllyasviel/Fooocus/blob/main/update_log.md)。
+
+## Fooocus
+
+**该章节来自 [Fooocus](https://github.com/lllyasviel/Fooocus) 项目。**
+
+Fooocus 是一个图像生成软件 (基于 [Gradio](https://www.gradio.app/))。
+
+Fooocus 是对于 Stable Diffusion 和 Midjourney 的重新思考以及设计：
+
+- 我们学习了 Stable Diffusion 的开源、免费、离线运行。
+
+- 我们学习了 Midjourney 的专注，不需要手动调整，专注于描述词以及图像。
+
+Fooocus 包含了许多内部优化以及质量改进。 忘记那些复杂困难的技术参数，享受人机交互带来的想象力的突破以及探索新的思维
+
+## Fooocus-API
+
+可能您已经尝试过通过 [Gradio 客户端](https://www.gradio.app/docs/client) 来接入 Fooocus，但您可能发现体验并不理想。
+
+Fooocus API 是基于 [FastAPI](https://fastapi.tiangolo.com/) 构建的一系列 `REST` 接口，它们使得利用 Fooocus 的强大功能变得简单易行。现在，您可以使用任何您喜欢的编程语言来轻松地与 Fooocus 进行交互。
+
+此外，我们还提供了详尽的 [API 文档](/docs/api_doc_zh.md) 和丰富的 [示例代码](/examples)，以���助您快速上手和深入了解如何有效地利用 Fooocus。
+
+# 开始
+
+## 在 Replicate 上运行
+
+现在你可以在 Replicate 上使用 Fooocus-API，在这儿： [konieshadow/fooocus-api](https://replicate.com/konieshadow/fooocus-api).
+
+使用预先调整参数的:
+
+- [konieshadow/fooocus-api-anime](https://replicate.com/konieshadow/fooocus-api-anime)
+- [konieshadow/fooocus-api-realistic](https://replicate.com/konieshadow/fooocus-api-realistic)
+
+我认为这是更简单的方法来体验 Fooocus 的强大
+
+> 出于某些原因，上述 replicate 上的实例版本无法更新，你可以参照 [push-a-model](https://replicate.com/docs/guides/push-a-model) 部署自己专用的实例。
+
+## 自托管
+
+需要 Python >= 3.10，或者使用 conda、venv 创建一个新的环境
+
+硬件需求来源于 Fooocus。 详细要求可以看[这里](https://github.com/lllyasviel/Fooocus#minimal-requirement)
+
+### conda
+
+按照下面的步骤启动一个 app：
+
+```shell
+conda env create -f environment.yaml
+conda activate fooocus-api
+```
+
+然后，执行 `python main.py` 启动 app ，默认情况下会监听在 `http://127.0.0.1:8888`
+
+> 如果是第一次运行，程序会自动处理完成剩余的环境配置、模型下载等工作，因此会等待一段时间。也可以预先配置好环境、下载模型，后面会提到。
+
+### venv
+
+和使用 conda 类似，创建虚拟环境，启动 app ，等待程序完成环境安装、模型下载
+
+```powershell
+# windows
+python -m venv venv
+.\venv\Scripts\Activate
+```
+
+```shell
+# linux
+python -m venv venv
+source venv/bin/activate
+```
+然后执行 `python main.py`
+
+### 预下载及安装
+
+如果想要手动配置环境以及放置模型，可以参考下面的步骤
+
+在创建完 conda 或者 venv 环境之后，按照下面的步骤手动配置环境、下载模型
+
+首先，安装 requirements： `pip install -r requirements.txt`
+
+然后安装 pytorch+cuda： `pip install torch==2.1.0 torchvision==0.16.0 torchaudio==2.1.0 --index-url https://download.pytorch.org/whl/cu121`
+
+更多安装信息在 pytorch 官方的 [previous-versions](https://pytorch.org/get-started/previous-versions/) 页面找到。
+
+> 关于 pytorch 和 cuda 的版本，Fooocus API 使用的是 Fooocus 推荐的版本，目前是 pytorch2.1.0+cuda12.1。如果你是个 "犟种" 非要用其他版本，我测试过也是可以的，不过启动的时候记得加上 `--skip-pip`，否则程序会自动替换为推荐版本。
+
+进入 `repositories` 的目录，下载的模型放到这个目录 `repositories\Fooocus\models`。如果你有一个已经安装完成的 Fooocus，在[这里](#已经有安装好的-fooocus)查看如何复用模型
+
+这里是一个启动必须下载的模型列表 (也可能不一样如果 [启动参数](#命令行参数) 不同的话):
+
+- checkpoint: 放到 `repositories\Fooocus\models\checkpoints`
+    + [juggernautXL_v8Rundiffusion.safetensors](https://huggingface.co/lllyasviel/fav_models/resolve/main/fav/juggernautXL_v8Rundiffusion.safetensors)
+
+- vae_approx: 放到 `repositories\Fooocus\models\vae_approx`
+    + [xlvaeapp.pth](https://huggingface.co/lllyasviel/misc/resolve/main/xlvaeapp.pth)
+    + [vaeapp_sd15.pth](https://huggingface.co/lllyasviel/misc/resolve/main/vaeapp_sd15.pt)
+    + [xl-to-v1_interposer-v3.1.safetensors](https://huggingface.co/lllyasviel/misc/resolve/main/xl-to-v1_interposer-v3.1.safetensors)
+
+- lora: 放到 `repositories\Fooocus\models\loras`
+    + [sd_xl_offset_example-lora_1.0.safetensors](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0/resolve/main/sd_xl_offset_example-lora_1.0.safetensors?download=true)
+
+> 国内不好下的到 [这儿](https://www.123pan.com/s/dF5A-SIQsh.html)下载， 提取码: `D4Mk`
+
+### 已经有安装好的 Fooocus
+
+如果你已经有一个安装好的且运行正常的 Fooocus， 推荐的方式是复用模型, 只需要将 Fooocus 根目录下的 `config.txt` 文件复制到 Fooocus API 的根目录即可。 查看 [Customization](https://github.com/lllyasviel/Fooocus#customization) 获取更多细节.
+
+使用这种方法 Fooocus 和 Fooocus API 会同时存在，独立运行互不干扰。
+
+> 不要将已安装的 Fooocus 目录复制到 repositories 目录。
+
+## 使用Docker启动
+
+开始之前，先安装 [NVIDIA Container Toolkit](https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/latest/install-guide.html)，这是 Docker 可以使用 GPU 的前提。
+
+运行
+
+```shell
+docker run -d --gpus=all \
+    -e NVIDIA_DRIVER_CAPABILITIES=compute,utility \
+    -e NVIDIA_VISIBLE_DEVICES=all \
+    -p 8888:8888 konieshadow/fooocus-api
+```
+
+一个更实用的例子:
+
+```shell
+mkdir ~/repositories
+mkdir -p ~/.cache/pip
+
+docker run -d --gpus=all \
+    -e NVIDIA_DRIVER_CAPABILITIES=compute,utility \
+    -e NVIDIA_VISIBLE_DEVICES=all \
+    -v ~/repositories:/app/repositories \
+    -v ~/.cache/pip:/root/.cache/pip \
+    -p 8888:8888 konieshadow/fooocus-api
+```
+
+这里把 `repositories` 和 `pip cache` 映射到了本地
+
+你还可以添加 `-e PIP_INDEX_URL={pypi-mirror-url}` 选项来更换 pip 源
+
+> 0.4.0.0 版本开始，镜像包含完整运行环境，因此只需要根据需要将 `models` 或者项目根目录进行映射即可
+> 比如：
+> ```
+> docker run -d --gpus all \
+>     -v /Fooocus-API:/app \
+>     -p 8888:8888 konieshadow/fooocus-api
+>```
+
+# 命令行参数
+
+- `-h, --help` 显示本帮助并退出
+- `--port PORT` 设置监听端口，默认：8888
+- `--host HOST` 设置监听地址，默认：127.0.0.1
+- `--base-url BASE_URL` 设置返回结果中的地址，默认是： http://host:port
+- `--log-level LOG_LEVEL` Uvicorn 中的日志等级，默认：info
+- `--skip-pip` 跳过启动时的 pip 安装
+- `--preload-pipeline` 启动 http server 之前加载 pipeline
+- `--queue-size QUEUE_SIZE` 工作队列大小，默认是 100 ，超过队列的请求会返回失败
+- `--queue-history QUEUE_HISTORY` 保留的作业历史，默认 0 即无限制，超过会被删除，包括生成的图像
+- `--webhook-url WEBHOOK_URL` 通知生成结果的 webhook 地址，默认为 None
+- `--persistent` 持久化历史记录到SQLite数据库，默认关闭
+- `--apikey APIKEY` 设置 apikey 以启用安全api，默认值：无
+
+从 v0.3.25 开始, Fooocus 的命令行选项也被支持，你可以在启动时加上 Fooocus 支持的选项
+
+比如(需要更大的显存):
+
+```
+python main.py --all-in-fp16 --always-gpu
+```
+
+完成的 Fooocus 命令行选项可以在[这儿](https://github.com/lllyasviel/Fooocus?tab=readme-ov-file#all-cmd-flags)找到。
+
+
+# 更新日志
+
+[CHANGELOG](./docs/change_logs_zh.md)
+
+更早的日志可以在 [release page](https://github.com/konieshadow/Fooocus-API/releases) 找到
+
+
+# Apis
+
+你可以在[这里](/docs/api_doc_zh.md)找到所有的 API 细节
+
+# License
+
+This repository is licensed under the [GUN General Public License v3.0](https://github.com/mrhan1993/Fooocus-API/blob/main/LICENSE)
+
+The default checkpoint is published by [RunDiffusion](https://huggingface.co/RunDiffusion), is licensed under the [CreativeML Open RAIL-M](https://github.com/mrhan1993/Fooocus-API/blob/main/CreativeMLOpenRAIL-M).
+
+or, you can find it [here](https://huggingface.co/spaces/CompVis/stable-diffusion-license)
+
+
+# 感谢 :purple_heart:
+
+感谢所有为改进 Fooocus API 做出贡献和努力的人。再次感谢 :sparkles: 社区万岁 :sparkles:!

cog.yaml

@@ -0,0 +1,48 @@
+# Configuration for Cog ⚙️
+# Reference: https://github.com/replicate/cog/blob/main/docs/yaml.md
+
+build:
+  # set to true if your model requires a GPU
+  gpu: true
+  cuda: "12.1"
+
+  # a list of ubuntu apt packages to install
+  system_packages:
+    - "libgl1-mesa-glx"
+    - "libglib2.0-0"
+
+  # python version in the form '3.11' or '3.11.4'
+  python_version: "3.11"
+
+  # a list of packages in the format <package-name>==<version>
+  python_packages:
+    - "torchsde==0.2.5"
+    - "einops==0.4.1"
+    - "transformers==4.30.2"
+    - "safetensors==0.3.1"
+    - "accelerate==0.21.0"
+    - "pyyaml==6.0"
+    - "Pillow==9.2.0"
+    - "scipy==1.9.3"
+    - "tqdm==4.64.1"
+    - "psutil==5.9.5"
+    - "pytorch_lightning==1.9.4"
+    - "omegaconf==2.2.3"
+    - "pygit2==1.12.2"
+    - "opencv-contrib-python==4.8.0.74"
+    - "onnxruntime==1.16.3"
+    - "timm==0.9.2"
+    - "torch==2.1.0"
+    - "torchvision==0.16.0"
+    - "colorlog==6.8.2"
+    - "rich==13.7.1"
+
+  # commands run after the environment is set up
+  # run:
+  #   - "echo env is ready!"
+  #   - "echo another command if needed"
+
+image: "r8.im/konieshadow/fooocus-api"
+
+# predict.py defines how predictions are run on your model
+predict: "predict.py:Predictor"

docs/api_doc_en.md

@@ -0,0 +1,983 @@
+- [Introduction](#introduction)
+- [Fooocus capability related interfaces](#fooocus-capability-related-interfaces)
+  - [text-to-image](#text-to-image)
+  - [image-upscale-vary](#image-upscale-vary)
+  - [image-inpaint-outpaint](#image-inpaint-outpaint)
+  - [image-prompt](#image-prompt)
+  - [text-to-image-with-image-prompt](#text-to-image-with-image-prompt)
+  - [describe](#describe)
+  - [all-models](#all-models)
+  - [refresh-models](#refresh-models)
+  - [styles](#styles)
+- [Fooocus API task related interfaces](#fooocus-api-task-related-interfaces)
+  - [job-queue](#job-queue)
+  - [query-job](#query-job)
+  - [job-history](#job-history)
+  - [stop](#stop)
+  - [ping](#ping)
+- [webhook](#webhook)
+- [public requests body](#public-requests-params)
+  - [AdvanceParams](#advanceparams)
+  - [lora](#lora)
+  - [response](#response)
+
+
+
+# Introduction
+
+Fooocus API are provided more than a dozen REST interfaces now, I roughly divide it into two categories, the first is the ability to call Fooocus, such as generating images, refreshing models, and so on, and the second is related to Fooocus API itself, mainly related to task queries. I will try to illustrate their role and usage and provide examples in the following content.
+
+> Almost all interface parameters have default values, which means you only need to send the parameters you are interested in. The complete parameters and default values can be viewed in the table.
+
+# Fooocus capability related interfaces
+
+## text-to-image
+
+Corresponding to the function of text to image in Fooocus
+
+**base info：**
+
+```yaml
+EndPoint: /v1/generation/text-to-image
+Method: Post
+DataType: json
+```
+**requests params：**
+
+| Name                    | Type           | Description                                                                                     |
+|-------------------------|----------------|-------------------------------------------------------------------------------------------------|
+| prompt                  | string         | prompt, default to empty string                                                                 |
+| negative_prompt         | string         | negative_prompt                                                                                 |
+| style_selections        | List[str]      | list of style, must be supported style, you can get all supported [style](#styles) here         |
+| performance_selection   | Enum           | performance_selection, must be one of `Speed`, `Quality`, `Extreme Speed` default to `Speed`    |
+| aspect_ratios_selection | str            | resolution, default to  `1152*896`                                                              |
+| image_number            | int            | the num of image to generate, default to 1 , max num is 32, note: Not a parallel interface      |
+| image_seed              | int            | seed, default to -1, meant random                                                               |
+| sharpness               | float          | sharpness, default to 2.0 , 0-30                                                                |
+| guidance_scale          | float          | guidance scale, default to 4.0 , 1-30                                                           |
+| base_model_name         | str            | base model, default to `juggernautXL_version6Rundiffusion.safetensors`                          |
+| refiner_model_name      | str            | refiner model, default to `None`                                                                |
+| refiner_switch          | float          | refiner switch, default to 0.5                                                                  |
+| loras                   | List[Lora]     | lora list, include conf, lora: [Lora](#lora)                                                    |
+| advanced_params         | AdvancedParams | Advanced params, [AdvancedParams](#advanceparams)                                               |
+| require_base64          | bool           | require base64, default to False                                                                |
+| save_meta               | bool           | save metadata to image, default True                                                            |
+| meta_scheme             | str            | metadata scheme, default 'fooocus', only support 'fooocus' now                                  |
+| save_extension          | str            | extension for saved image, default 'png'                                                        |
+| save_name               | str            | image name saved, default job_id + seq                                                          |
+| read_wildcards_in_order | bool           | read wildcards in order, default False                                                          |
+| async_process           | bool           | is async, default to False                                                                      |
+| webhook_url             | str            | after async task completed, address for callback, default to None, refer to [webhook](#webhook) |
+
+**response params：**
+
+Most response have the same structure, but different parts will be specifically explained 
+
+This interface returns a universal response structure, refer to [response](#response)
+
+**request example：**
+
+```python
+host = "http://127.0.0.1:8888"
+
+def text2img(params: dict) -> dict:
+    """
+    text to image
+    """
+    result = requests.post(url=f"{host}/v1/generation/text-to-image",
+                           data=json.dumps(params),
+                           headers={"Content-Type": "application/json"})
+    return result.json()
+
+result =text2img({
+    "prompt": "1girl sitting on the ground",
+    "async_process": True})
+print(result)
+```
+
+## image-upscale-vary
+
+Corresponding to the function of Upscale or Variation in Fooocus
+
+the requests body for this interface based on [text-to-image](#text-to-image), so I will only list the difference with [text-to-image](#text-to-image)
+
+In addition, the interface provides two versions, and there is no functional difference between the two versions, mainly due to slight differences in request methods
+
+**base info：**
+
+```yaml
+EndPoint_V1: /v1/generation/image-upscale-vary
+EndPoint_V2: /v2/generation/image-upscale-vary
+Method: Post
+DataType: form|json
+```
+
+### V1
+
+**requests params**
+
+| Name             | Type                | Description                                                                                                                               |
+|------------------|---------------------|-------------------------------------------------------------------------------------------------------------------------------------------|
+| input_image      | string($binary)     | binary image                                                                                                                              |
+| uov_method       | Enum                | 'Vary (Subtle)','Vary (Strong)','Upscale (1.5x)','Upscale (2x)','Upscale (Fast 2x)','Upscale (Custom)'                                    |
+| upscale_value    | float               | default to None , 1.0-5.0, magnification, only for uov_method is 'Upscale (Custom)'                                                       |
+| style_selections | List[str]           | list Fooocus style seg with comma                                                                                                         |
+| loras            | str(List[Lora])     | list for lora, with configure, lora: [Lora](#lora), example: [{"model_name": "sd_xl_offset_example-lora_1.0.safetensors", "weight": 0.5}] |
+| advanced_params  | str(AdvancedParams) | AdvancedParams, AdvancedParams: [AdvancedParams](#advanceparams), send with str, None is available                                        |
+
+**response params：**
+
+This interface returns a universal response structure, refer to [response](#response)
+
+**requests example：**
+
+```python
+# headers should not contain {"Content-Type": "application/json"}
+
+host = "http://127.0.0.1:8888"
+image = open("./examples/imgs/bear.jpg", "rb").read()
+
+def upscale_vary(image, params: dict) -> dict:
+    """
+    Upscale or Vary
+    """
+    response = requests.post(url=f"{host}/v1/generation/image-upscale-vary",
+                        data=params,
+                        files={"input_image": image})
+    return response.json()
+
+result =upscale_vary(image=image,
+                     params={
+                         "uov_method": "Upscale (2x)",
+                         "async_process": True
+                     })
+print(json.dumps(result, indent=4, ensure_ascii=False))
+```
+
+### V2
+
+**requests params**
+
+| Name          | Type                | Description                                                                                                                           |
+|---------------|---------------------|---------------------------------------------------------------------------------------------------------------------------------------|
+| uov_method    | UpscaleOrVaryMethod | Enum type, value should one of 'Vary (Subtle)','Vary (Strong)','Upscale (1.5x)','Upscale (2x)','Upscale (Fast 2x)','Upscale (Custom)' |
+| upscale_value | float               | default to None , 1.0-5.0, magnification, only for uov_method is 'Upscale (Custom)'                                                   |
+| input_image   | str                 | input image, base64 str, or a URL                                                                                                     |
+
+**response params：**
+
+This interface returns a universal response structure, refer to [response](#response)
+
+**requests params：**
+
+```python
+host = "http://127.0.0.1:8888"
+image = open("./examples/imgs/bear.jpg", "rb").read()
+
+def upscale_vary(image, params: dict) -> dict:
+    """
+    Upscale or Vary
+    """
+    params["input_image"] = base64.b64encode(image).decode('utf-8') 
+    response = requests.post(url=f"{host}/v2/generation/image-upscale-vary",
+                        data=json.dumps(params),
+                        headers={"Content-Type": "application/json"},
+                        timeout=300)
+    return response.json()
+
+result =upscale_vary(image=image,
+                     params={
+                         "uov_method": "Upscale (2x)",
+                         "async_process": True
+                     })
+print(json.dumps(result, indent=4, ensure_ascii=False))
+```
+
+## image-inpaint-outpaint
+
+**base info：**
+
+```yaml
+EndPoint_V1: /v1/generation/image-inpaint-outpaint
+EndPoint_V2: /v2/generation/image-inpaint-outpaint
+Method: Post
+DataType: form|json
+```
+
+### V1
+
+**requests params**
+
+| Name                      | Type                | Description                                                                                                                      |
+|---------------------------|---------------------|----------------------------------------------------------------------------------------------------------------------------------|
+| input_image               | string($binary)     | binary image                                                                                                                     |
+| input_mask                | string($binary)     | binary image                                                                                                                     |
+| inpaint_additional_prompt | string              | additional_prompt                                                                                                                |
+| outpaint_selections       | str                 | Image extension direction , 'Left', 'Right', 'Top', 'Bottom' seg with comma                                                      |
+| outpaint_distance_left    | int                 | Image extension distance, default to 0                                                                                           |
+| outpaint_distance_right   | int                 | Image extension distance, default to 0                                                                                           |
+| outpaint_distance_top     | int                 | Image extension distance, default to 0                                                                                           |
+| outpaint_distance_bottom  | int                 | Image extension distance, default to 0                                                                                           |
+| style_selections          | List[str]           | list Fooocus style seg with comma                                                                                                |
+| loras                     | str(List[Lora])     | list for lora, with configure, lora: Lora, example: [{"model_name": "sd_xl_offset_example-lora_1.0.safetensors", "weight": 0.5}] |
+| advanced_params           | str(AdvancedParams) | AdvancedParams, AdvancedParams: AdvancedParams, send with str, None is available                                                 |
+
+**response params：**
+
+This interface returns a universal response structure, refer to [response](#response)
+
+**requests example：**
+
+```python
+# example for inpaint outpaint v1
+host = "http://127.0.0.1:8888"
+image = open("./examples/imgs/bear.jpg", "rb").read()
+
+def inpaint_outpaint(params: dict, input_image: bytes, input_mask: bytes = None) -> dict:
+    """
+    example for inpaint outpaint v1
+    """
+    response = requests.post(url=f"{host}/v1/generation/image-inpaint-outpaint",
+                        data=params,
+                        files={"input_image": input_image,
+                               "input_mask": input_mask})
+    return response.json()
+
+# image extension example
+result = inpaint_outpaint(params={
+                            "outpaint_selections": "Left,Right",
+                            "async_process": True},
+                          input_image=image,
+                          input_mask=None)
+print(json.dumps(result, indent=4, ensure_ascii=False))
+
+# image inpaint example
+source = open("./examples/imgs/s.jpg", "rb").read()
+mask = open("./examples/imgs/m.png", "rb").read()
+result = inpaint_outpaint(params={
+                            "prompt": "a cat",
+                            "async_process": True},
+                          input_image=source,
+                          input_mask=mask)
+print(json.dumps(result, indent=4, ensure_ascii=False))
+```
+
+### V2
+
+**requests params**
+
+| Name                      | Type                    | Description                                                                     |
+|---------------------------|-------------------------|---------------------------------------------------------------------------------|
+| input_image               | str                     | input image, base64 str, or a URL                                               |
+| input_mask                | str                     | input mask, base64 str, or a URL                                                |
+| inpaint_additional_prompt | str                     | additional prompt                                                               |
+| outpaint_selections       | List[OutpaintExpansion] | OutpaintExpansion is Enum, value should one of "Left", "Right", "Top", "Bottom" |
+| outpaint_distance_left    | int                     | Image extension distance, default to 0                                          |
+| outpaint_distance_right   | int                     | Image extension distance, default to 0                                          |
+| outpaint_distance_top     | int                     | Image extension distance, default to 0                                          |
+| outpaint_distance_bottom  | int                     | Image extension distance, default to 0                                          |
+
+**response params：**
+
+This interface returns a universal response structure, refer to [response](#response)[response params](#response)
+
+**requests example：**
+
+```python
+# example for inpaint outpaint v2
+host = "http://127.0.0.1:8888"
+image = open("./examples/imgs/bear.jpg", "rb").read()
+
+def inpaint_outpaint(params: dict) -> dict:
+    """
+    example for inpaint outpaint v2
+    """
+    response = requests.post(url=f"{host}/v2/generation/image-inpaint-outpaint",
+                        data=json.dumps(params),
+                        headers={"Content-Type": "application/json"})
+    return response.json()
+
+# image extension example
+result = inpaint_outpaint(params={
+                            "input_image": base64.b64encode(image).decode('utf-8'),
+                            "input_mask": None,
+                            "outpaint_selections": ["Left", "Right"],
+                            "async_process": True})
+print(json.dumps(result, indent=4, ensure_ascii=False))
+
+# image inpaint example
+source = open("./examples/imgs/s.jpg", "rb").read()
+mask = open("./examples/imgs/m.png", "rb").read()
+result = inpaint_outpaint(params={
+                            "prompt": "a cat",
+                            "input_image": base64.b64encode(source).decode('utf-8'),
+                            "input_mask": base64.b64encode(mask).decode('utf-8'),
+                            "async_process": True})
+print(json.dumps(result, indent=4, ensure_ascii=False))
+```
+
+## image-prompt
+
+`v0.3.27` has a break change. Interface based on change to [inpaint-outpaint](#image-inpaint-outpaint)
+
+after v0.3.27, this interface implements the functions of `inpaint_outpaint` and `image-prompt`.
+
+>  Multi-function interface, which does not implement the functions of `inpaint_outpaint` and `image-prompt` at the same time in the same request
+
+**base info：**
+
+```yaml
+EndPoint_V1: /v1/generation/image-prompt
+EndPoint_V2: /v2/generation/image-prompt
+Method: Post
+DataType: form|json
+```
+
+### V1
+
+**requests params**
+
+| Name                      | Type                | Description                                                                                                                      |
+|---------------------------|---------------------|----------------------------------------------------------------------------------------------------------------------------------|
+| input_image               | Bytes               | binary image, use for inpaint                                                                                                    |
+| input_mask                | Bytes               | binary image mask, use for inpaint                                                                                               |
+| inpaint_additional_prompt | str                 | inpaint additional prompt                                                                                                        |
+| outpaint_selections       | str                 | Image extension direction , 'Left', 'Right', 'Top', 'Bottom' seg with comma                                                      |
+| outpaint_distance_left    | int                 | Image extension distance, default to 0                                                                                           |
+| outpaint_distance_right   | int                 | Image extension distance, default to 0                                                                                           |
+| outpaint_distance_top     | int                 | Image extension distance, default to 0                                                                                           |
+| outpaint_distance_bottom  | int                 | Image extension distance, default to 0                                                                                           |
+| cn_img1                   | string($binary)     | binary image                                                                                                                     |
+| cn_stop1                  | float               | default to 0.6                                                                                                                   |
+| cn_weight1                | float               | default to 0.6                                                                                                                   |
+| cn_type1                  | Enum                | should one of "ImagePrompt", "FaceSwap", "PyraCanny", "CPDS"                                                                     |
+| cn_img2                   | string($binary)     | binary image                                                                                                                     |
+| cn_stop2                  | float               | default to 0.6                                                                                                                   |
+| cn_weight2                | float               | default to 0.6                                                                                                                   |
+| cn_type2                  | Enum                | should one of "ImagePrompt", "FaceSwap", "PyraCanny", "CPDS"                                                                     |
+| cn_img3                   | string($binary)     | binary image                                                                                                                     |
+| cn_stop3                  | float               | default to 0.6                                                                                                                   |
+| cn_weight3                | float               | default to 0.6                                                                                                                   |
+| cn_type3                  | Enum                | should one of "ImagePrompt", "FaceSwap", "PyraCanny", "CPDS"                                                                     |
+| cn_img4                   | string($binary)     | binary image                                                                                                                     |
+| cn_stop4                  | float               | default to 0.6                                                                                                                   |
+| cn_weight4                | float               | default to 0.6                                                                                                                   |
+| cn_type4                  | Enum                | should one of "ImagePrompt", "FaceSwap", "PyraCanny", "CPDS"                                                                     |
+| style_selections          | List[str]           | list Fooocus style seg with comma                                                                                                |
+| loras                     | str(List[Lora])     | list for lora, with configure, lora: Lora, example: [{"model_name": "sd_xl_offset_example-lora_1.0.safetensors", "weight": 0.5}] |
+| advanced_params           | str(AdvancedParams) | AdvancedParams, AdvancedParams: AdvancedParams, send with str, None is available                                                 |
+
+**response params：**
+
+This interface returns a universal response structure, refer to [response](#response)[response params](#response)
+
+**requests example：**
+
+```python
+# image_prompt v1 example
+host = "http://127.0.0.1:8888"
+image = open("./examples/imgs/bear.jpg", "rb").read()
+source = open("./examples/imgs/s.jpg", "rb").read()
+mask = open("./examples/imgs/m.png", "rb").read()
+
+def image_prompt(params: dict,
+                 input_image: bytes=None,
+                 input_mask: bytes=None,
+                 cn_img1: bytes=None,
+                 cn_img2: bytes=None,
+                 cn_img3: bytes=None,
+                 cn_img4: bytes=None,) -> dict:
+    """
+    image prompt
+    """
+    response = requests.post(url=f"{host}/v1/generation/image-prompt",
+                             data=params,
+                             files={
+                                 "input_image": input_image,
+                                 "input_mask": input_mask,
+                                 "cn_img1": cn_img1,
+                                 "cn_img2": cn_img2,
+                                 "cn_img3": cn_img3,
+                                 "cn_img4": cn_img4,
+                              })
+    return response.json()
+
+# image extend
+params = {
+    "outpaint_selections": ["Left", "Right"],
+    "image_prompts": [] # required, can be empty list
+}
+result = image_prompt(params=params, input_image=image)
+print(json.dumps(result, indent=4, ensure_ascii=False))
+
+# inpaint
+
+params = {
+    "prompt": "1girl sitting on the chair",
+    "image_prompts": [], # required, can be empty list
+    "async_process": True
+}
+result = image_prompt(params=params, input_image=source, input_mask=mask)
+print(json.dumps(result, indent=4, ensure_ascii=False))
+
+# image prompt
+
+params = {
+    "prompt": "1girl sitting on the chair",
+    "image_prompts": [
+        {
+            "cn_stop": 0.6,
+            "cn_weight": 0.6,
+            "cn_type": "ImagePrompt"
+        },{
+            "cn_stop": 0.6,
+            "cn_weight": 0.6,
+            "cn_type": "ImagePrompt"
+        }]
+    }
+result = image_prompt(params=params, cn_img1=image, cn_img2=source)
+print(json.dumps(result, indent=4, ensure_ascii=False))
+```
+
+### V2
+
+**requests params**
+
+| Name                      | Type              | Description                                                                 |
+|---------------------------|-------------------|-----------------------------------------------------------------------------|
+| input_image               | str               | base64 image, or a URL, use for inpaint                                     |
+| input_mask                | str               | base64 image mask, or a URL, use for inpaint                                |
+| inpaint_additional_prompt | str               | inpaint additional prompt                                                   |
+| outpaint_selections       | List[]            | Image extension direction , 'Left', 'Right', 'Top', 'Bottom' seg with comma |
+| outpaint_distance_left    | int               | Image extension distance, default to 0                                      |
+| outpaint_distance_right   | int               | Image extension distance, default to 0                                      |
+| outpaint_distance_top     | int               | Image extension distance, default to 0                                      |
+| outpaint_distance_bottom  | int               | Image extension distance, default to 0                                      |
+| image_prompts             | List[ImagePrompt] | image list, include config, ImagePrompt struct：                             |
+
+**ImagePrompt**
+
+| Name      | Type           | Description                                                                       |
+|-----------|----------------|-----------------------------------------------------------------------------------|
+| cn_img    | str            | input image, base64 str, or a URL                                                 |
+| cn_stop   | float          | 0-1, default to 0.5                                                               |
+| cn_weight | float          | weight, 0-2, default to 1.0                                                       |
+| cn_type   | ControlNetType | ControlNetType Enum, should one of "ImagePrompt", "FaceSwap", "PyraCanny", "CPDS" |
+
+**response params：**
+
+This interface returns a universal response structure, refer to [response](#response)[response params](#response)
+
+**requests example：**
+
+```python
+# image_prompt v2 example
+host = "http://127.0.0.1:8888"
+image = open("./examples/imgs/bear.jpg", "rb").read()
+source = open("./examples/imgs/s.jpg", "rb").read()
+mask = open("./examples/imgs/m.png", "rb").read()
+
+def image_prompt(params: dict) -> dict:
+    """
+    image prompt
+    """
+    response = requests.post(url=f"{host}/v2/generation/image-prompt",
+                             data=json.dumps(params),
+                             headers={"Content-Type": "application/json"})
+    return response.json()
+
+# image extend
+params = {
+    "input_image": base64.b64encode(image).decode('utf-8'),
+    "outpaint_selections": ["Left", "Right"],
+    "image_prompts": [] # required, can be empty list
+}
+result = image_prompt(params)
+print(json.dumps(result, indent=4, ensure_ascii=False))
+
+# inpaint
+
+params = {
+    "prompt": "1girl sitting on the chair",
+    "input_image": base64.b64encode(source).decode('utf-8'),
+    "input_mask": base64.b64encode(mask).decode('utf-8'),
+    "image_prompts": [], # required, can be empty list
+    "async_process": True
+}
+result = image_prompt(params)
+print(json.dumps(result, indent=4, ensure_ascii=False))
+
+# image prompt
+
+params = {
+    "prompt": "1girl sitting on the chair",
+    "image_prompts": [
+        {
+            "cn_img": base64.b64encode(source).decode('utf-8'),
+            "cn_stop": 0.6,
+            "cn_weight": 0.6,
+            "cn_type": "ImagePrompt"
+        },{
+            "cn_img": base64.b64encode(image).decode('utf-8'),
+            "cn_stop": 0.6,
+            "cn_weight": 0.6,
+            "cn_type": "ImagePrompt"
+        }]
+    }
+result = image_prompt(params)
+print(json.dumps(result, indent=4, ensure_ascii=False))
+```
+
+## text to image with image prompt
+
+this interface only provides v2 version
+
+**base info:**
+
+```yaml
+EndPoint: /v2/generation/text-to-image-with-ip
+Method: Post
+DataType: json
+```
+
+**requests params**
+
+| Name          | Type              | Description |
+|---------------|-------------------|-------------|
+| image_prompts | List[ImagePrompt] | Image list  |
+
+**requests example**:
+
+```python
+# text to image with image prompt example
+host = "http://127.0.0.1:8888"
+image = open("./examples/imgs/bear.jpg", "rb").read()
+source = open("./examples/imgs/s.jpg", "rb").read()
+def image_prompt(params: dict) -> dict:
+    """
+    image prompt
+    """
+    response = requests.post(url=f"{host}/v2/generation/text-to-image-with-ip",
+                             data=json.dumps(params),
+                             headers={"Content-Type": "application/json"})
+    return response.json()
+
+params = {
+    "prompt": "A bear",
+    "image_prompts": [
+        {
+            "cn_img": base64.b64encode(source).decode('utf-8'),
+            "cn_stop": 0.6,
+            "cn_weight": 0.6,
+            "cn_type": "ImagePrompt"
+        },{
+            "cn_img": base64.b64encode(image).decode('utf-8'),
+            "cn_stop": 0.6,
+            "cn_weight": 0.6,
+            "cn_type": "ImagePrompt"
+        }
+    ]
+}
+result = image_prompt(params)
+print(json.dumps(result, indent=4, ensure_ascii=False))
+```
+
+## describe
+
+**base info：**
+
+```yaml
+EndPoint: /v1/tools/describe-image
+Method: Post
+DataType: form
+```
+
+**requests params**
+
+| Name | Type | Description                              |
+|------|------|------------------------------------------|
+| type | Enum | type, should be one of "Photo", "Anime"  |
+
+**requests example**:
+
+```python
+def describe_image(image: bytes,
+                   params: dict = {"type": "Photo"}) -> dict:
+    """
+    describe-image
+    """
+    response = requests.post(url="http://127.0.0.1:8888/v1/tools/describe-image",
+                        params=params,
+                        files={
+                            "image": image
+                        },
+                        timeout=30)
+    return response.json()
+```
+
+**response example**:
+
+```python
+{
+  "describe": "a young woman posing with her hands behind her head"
+}
+```
+
+--------------------------------------------
+
+## all-models
+
+**base info：**
+
+```yaml
+EndPoint: /v1/engines/all-models
+Method: Get
+```
+
+**requests example**:
+
+```python
+def all_models() -> dict:
+    """
+    all-models
+    """
+    response = requests.get(url="http://127.0.0.1:8888/v1/engines/all-models",
+                        timeout=30)
+    return response.json()
+```
+
+**response params**:
+
+```python
+{
+  "model_filenames": [
+    "juggernautXL_version6Rundiffusion.safetensors",
+    "sd_xl_base_1.0_0.9vae.safetensors",
+    "sd_xl_refiner_1.0_0.9vae.safetensors"
+  ],
+  "lora_filenames": [
+    "sd_xl_offset_example-lora_1.0.safetensors"
+  ]
+}
+```
+
+## refresh-models
+
+**base info：**
+
+```yaml
+EndPoint: /v1/engines/refresh-models
+Method: Post
+```
+
+> Removed, use [all-models](#all-models) instead
+
+**requests example**
+```python
+def refresh() -> dict:
+    """
+    refresh-models
+    """
+    response = requests.post(url="http://127.0.0.1:8888/v1/engines/refresh-models",
+                        timeout=30)
+    return response.json()
+```
+
+**response params**
+```python
+{
+  "model_filenames": [
+    "juggernautXL_version6Rundiffusion.safetensors",
+    "sd_xl_base_1.0_0.9vae.safetensors",
+    "sd_xl_refiner_1.0_0.9vae.safetensors"
+  ],
+  "lora_filenames": [
+    "sd_xl_offset_example-lora_1.0.safetensors"
+  ]
+}
+```
+
+## styles
+
+**base info：**
+
+```yaml
+EndPoint: /v1/engines/styles
+Method: Get
+```
+
+**requests example**:
+
+```python
+def styles() -> dict:
+    """
+    styles
+    """
+    response = requests.get(url="http://127.0.0.1:8888/v1/engines/styles",
+                        timeout=30)
+    return response.json()
+```
+
+**response params**:
+
+```python
+[
+  "Fooocus V2",
+  "Fooocus Enhance",
+  ...
+  "Watercolor 2",
+  "Whimsical And Playful"
+]
+```
+
+# Fooocus API task related interfaces
+
+## job-queue
+
+**base info：**
+
+```yaml
+EndPoint: /v1/engines/job-queue
+Method: Get
+```
+
+**requests example**:
+
+```python
+def job_queue() -> dict:
+    """
+    job-queue
+    """
+    response = requests.get(url="http://127.0.0.1:8888/v1/generation/job-queue",
+                        timeout=30)
+    return response.json()
+```
+
+**response params**:
+
+```python
+{
+  "running_size": 0,
+  "finished_size": 1,
+  "last_job_id": "cac3914a-926d-4b6f-a46a-83794a0ce1d4"
+}
+```
+
+## query-job
+
+**base info：**
+
+```yaml
+EndPoint: /v1/generation/query-job
+Method: Get
+```
+
+**requests example**:
+```python
+def taskResult(task_id: str) -> dict:
+    # get task status
+    task_status = requests.get(url="http://127.0.0.1:8888/v1/generation/query-job",
+                               params={"job_id": task_id,
+                                       "require_step_preview": False},
+                               timeout=30)
+
+    return task_status.json()
+```
+
+**response params**:
+```python
+{
+  "job_id": "cac3914a-926d-4b6f-a46a-83794a0ce1d4",
+  "job_type": "Text to Image",
+  "job_stage": "SUCCESS",
+  "job_progress": 100,
+  "job_status": "Finished",
+  "job_step_preview": null,
+  "job_result": [
+    {
+      "base64": null,
+      "url": "http://127.0.0.1:8888/files/2023-11-27/b928e50e-3c09-4187-a3f9-1c12280bfd95.png",
+      "seed": 8228839561385006000,
+      "finish_reason": "SUCCESS"
+    }
+  ]
+}
+```
+
+## job-history
+
+**base info：**
+
+```yaml
+EndPoint: /v1/generation/job-history
+Method: get
+```
+
+**requests example**:
+
+```python
+def job-history() -> dict:
+    """
+    job-history
+    """
+    response = requests.get(url="http://127.0.0.1:8888/v1/generation/job-history",
+                        timeout=30)
+    return response.json()
+```
+
+**response params**:
+
+```python
+{
+  "queue": [],
+  "history": [
+    "job_id": "cac3914a-926d-4b6f-a46a-83794a0ce1d4",
+    "is_finished": True
+  ]
+}
+```
+
+## stop
+
+**base info：**
+
+```yaml
+EndPoint: /v1/generation/stop
+Method: post
+```
+
+**requests example**:
+
+```python
+def stop() -> dict:
+    """
+    stop
+    """
+    response = requests.post(url="http://127.0.0.1:8888/v1/generation/stop",
+                        timeout=30)
+    return response.json()
+```
+
+**response params**:
+
+```python
+{
+  "msg": "success"
+}
+```
+
+## ping
+
+**base info：**
+
+```yaml
+EndPoint: /ping
+Method: get
+```
+
+pong
+
+# webhook
+
+You can specify an address through '--webhook_url' on the command line so that you can receive notifications after asynchronous tasks are completed
+
+Here is a simple example to demonstrate how 'webhook' works
+
+First，start a simple server using the following code:
+
+```python
+from fastapi import FastAPI
+import uvicorn
+
+app = FastAPI()
+
+@app.post("/status")
+async def status(requests: dict):
+    print(requests)
+
+uvicorn.run(app, host="0.0.0.0", port=8000)
+```
+
+Then, start Fooocus API with `--webhook-url http://host:8000/status`
+
+Submit a task in any way, and after completion, you will see the task completion information in the background of this simple server：
+
+```python
+{'job_id': '717ec0b5-85df-4174-80d6-bddf93cd8248', 'job_result': [{'url': 'http://127.0.0.1:8888/files/2023-12-29/f1eca704-718e-4781-9d5f-82d41aa799d7.png', 'seed': '3283449865282320931'}]}
+```
+
+# public requests params
+
+## AdvanceParams
+
+| Name                                 | Type  | Description                                                                      |
+|--------------------------------------|-------|----------------------------------------------------------------------------------|
+| disable_preview                      | bool  | disable preview, default to False                                                |
+| adm_scaler_positive                  | float | ADM Guidance Scaler, default to 1.5, range 0.1-3.0                               |
+| adm_scaler_negative                  | float | negative ADM Guidance Scaler, default to 0.8, range 0.1-3.0                      |
+| adm_scaler_end                       | float | ADM Guidance Scaler end value, default to 0.5, range 0.0-1.0                     |
+| refiner_swap_method                  | str   | refiner model swap method, default to `joint`                                    |
+| adaptive_cfg                         | float | CFG Mimicking from TSNR, default to 7.0, range 1.0-30.0                          |
+| sampler_name                         | str   | sampler, default to `default_sampler`                                            |
+| scheduler_name                       | str   | scheduler, default to `default_scheduler`                                        |
+| overwrite_step                       | int   | Forced Overwrite of Sampling Step, default to -1, range -1-200                   |
+| overwrite_switch                     | int   | Forced Overwrite of Refiner Switch Step, default to -1, range -1-200             |
+| overwrite_width                      | int   | Forced Overwrite of Generating Width, default to -1, range -1-2048               |
+| overwrite_height                     | int   | Forced Overwrite of Generating Height, default to -1, range -1-2048              |
+| overwrite_vary_strength              | float | Forced Overwrite of Denoising Strength of "Vary", default to -1, range -1-1.0    |
+| overwrite_upscale_strength           | float | Forced Overwrite of Denoising Strength of "Upscale", default to -1, range -1-1.0 |
+| mixing_image_prompt_and_vary_upscale | bool  | Mixing Image Prompt and Vary/Upscale, default to False                           |
+| mixing_image_prompt_and_inpaint      | bool  | Mixing Image Prompt and Inpaint, default to False                                |
+| debugging_cn_preprocessor            | bool  | Debug Preprocessors, default to False                                            |
+| skipping_cn_preprocessor             | bool  | Skip Preprocessors, default to False                                             |
+| controlnet_softness                  | float | Softness of ControlNet, default to 0.25, range 0.0-1.0                           |
+| canny_low_threshold                  | int   | Canny Low Threshold, default to 64, range 1-255                                  |
+| canny_high_threshold                 | int   | Canny High Threshold, default to 128, range 1-255                                |
+| freeu_enabled                        | bool  | FreeU enabled, default to False                                                  |
+| freeu_b1                             | float | FreeU B1, default to 1.01                                                        |
+| freeu_b2                             | float | FreeU B2, default to 1.02                                                        |
+| freeu_s1                             | float | FreeU B3, default to 0.99                                                        |
+| freeu_s2                             | float | FreeU B4, default to 0.95                                                        |
+| debugging_inpaint_preprocessor       | bool  | Debug Inpaint Preprocessing, default to False                                    |
+| inpaint_disable_initial_latent       | bool  | Disable initial latent in inpaint, default to False                              |
+| inpaint_engine                       | str   | Inpaint Engine, default to `v2.6`                                                |
+| inpaint_strength                     | float | Inpaint Denoising Strength, default to 1.0, range 0.0-1.0                        |
+| inpaint_respective_field             | float | Inpaint Respective Field, default to 1.0, range 0.0-1.0                          |
+| inpaint_mask_upload_checkbox         | bool  | upload mask, default False                                                       |
+| invert_mask_checkbox                 | bool  | revert mask, default False                                                       |
+| inpaint_erode_or_dilate              | int   | Mask Erode or Dilate, default 0, -64-64                                          |
+
+## lora
+
+| Name       | Type  | Description            |
+|------------|-------|------------------------|
+| enabled    | bool  | enable lora            |
+| model_name | str   | model name             |
+| weight     | float | weight, default to 0.5 |
+
+## response
+
+success response：
+
+**async_process: True**
+
+| Name             | Type  | Description  |
+|------------------|-------|--------------|
+| job_id           | int   | job ID       |
+| job_type         | str   | job type     |
+| job_stage        | str   | job stage    |
+| job_progress     | float | job progress |
+| job_status       | str   | job status   |
+| job_step_preview | str   | job preview  |
+| job_result       | str   | job result   |
+
+**async_process: False**
+
+| Name          | Type | Description                                                                      |
+|---------------|------|----------------------------------------------------------------------------------|
+| base64        | str  | base64 image, according to `require_base64` params determines whether it is null |
+| url           | str  | result image url                                                                 |
+| seed          | int  | image seed                                                                       |
+| finish_reason | str  | finish reason                                                                    |
+
+fail response：

docs/api_doc_zh.md

@@ -0,0 +1,987 @@
+- [简介](#简介)
+- [Fooocus 能力相关接口](#fooocus-能力相关接口)
+  - [文生图 | text-to-image](#文生图--text-to-image)
+  - [图像放大 | image-upscale-vary](#图像放大--image-upscale-vary)
+  - [局部重绘 | image-inpaint-outpaint](#局部重绘--image-inpaint-outpaint)
+  - [图生图 | image-prompt](#图生图--image-prompt)
+  - [text-to-image-with-image prompt](#text-to-image-with-image-prompt)
+  - [图像反推 | describe](#图像反推--describe)
+  - [列出模型 | all-models](#列出模型--all-models)
+  - [刷新模型 | refresh-models](#刷新模型--refresh-models)
+  - [样式 | styles](#样式--styles)
+- [Fooocus API 任务相关接口](#fooocus-api-任务相关接口)
+  - [任务队列 | job-queue](#任务队列--job-queue)
+  - [查询任务 | query-job](#查询任务--query-job)
+  - [查询任务历史 | job-history](#查询任务历史--job-history)
+  - [停止任务 | stop](#停止任务--stop)
+  - [ping](#ping)
+- [webhook](#webhook)
+- [公共请求体](#公共请求体)
+  - [高级参数 | AdvanceParams](#高级参数--advanceparams)
+  - [lora](#lora)
+  - [响应参数 | response](#响应参数--response)
+
+
+
+# 简介
+
+Fooocus API 目前提供了十多个 REST 接口, 我大致将其分为两类, 第一类用来调用 Fooocus 的能力, 比如生成图像、刷新模型之类的, 第二类为 Fooocus API 自身相关的, 主要是任务查询相关。我会在接下来的内容中尝试说明它们的作用以及用法并提供示例。
+
+> 几乎所有的接口参数都有默认值，这意味着你只需要发送你感兴趣的参数即可。完整的参数以及默认值可以通过表格查看
+
+# Fooocus 能力相关接口
+
+## 文生图 | text-to-image
+
+对应 Fooocus 中的文生图功能
+
+**基础信息：**
+
+```yaml
+EndPoint: /v1/generation/text-to-image
+Method: Post
+DataType: json
+```
+**请求参数：**
+
+| Name                    | Type           | Description                                                             |
+|-------------------------|----------------|-------------------------------------------------------------------------|
+| prompt                  | string         | 描述词, 默认为空字符串                                                            |
+| negative_prompt         | string         | 描述词, 反向描述词                                                              |
+| style_selections        | List[str]      | 风格列表, 需要是受支持的风格, 可以通过 [样式接口](#样式--styles) 获取所有支持的样式                     |
+| performance_selection   | Enum           | 性能选择, `Speed`, `Quality`, `Extreme Speed`, `Lightning` 中的一个, 默认 `Speed` |
+| aspect_ratios_selection | str            | 分辨率, 默认 '1152*896'                                                      |
+| image_number            | int            | 生成图片数量, 默认 1 , 最大32, 注: 非并行接口                                           |
+| image_seed              | int            | 图片种子, 默认 -1, 即随机生成                                                      |
+| sharpness               | float          | 锐度, 默认 2.0 , 0-30                                                       |
+| guidance_scale          | float          | 引导比例, 默认 4.0 , 1-30                                                     |
+| base_model_name         | str            | 基础模型, 默认 `juggernautXL_version6Rundiffusion.safetensors`                |
+| refiner_model_name      | str            | 优化模型, 默认 `None`                                                         |
+| refiner_switch          | float          | 优化模型切换时机, 默认 0.5                                                        |
+| loras                   | List[Lora]     | lora 模型列表, 包含配置, lora 结构: [Lora](#lora)                                 |
+| advanced_params         | AdvancedParams | 高级参数, AdvancedParams 结构 [AdvancedParams](#高级参数--advanceparams)          |
+| save_meta               | bool           | 是否保存元数据, 默认 True                                                        |
+| meta_scheme             | str            | 元数据方案, 默认 'fooocus', 目前只支持 'fooocus'                                    |
+| save_extension          | str            | 保存文件扩展名, 默认 'png'                                                       |
+| save_name               | str            | 保存文件名, 默认 job_id + 序号                                                   |
+| read_wildcards_in_order | bool           | 是否按照顺序读取通配符, 默认 False                                                   |
+| require_base64          | bool           | 是否返回base64编码, 默认 False                                                  |
+| async_process           | bool           | 是否异步处理, 默认 False                                                        |
+| webhook_url             | str            | 异步处理完成后, 触发的 webhook 地址, 参考[webhook](#webhook)                          |
+
+**响应参数：**
+
+多数响应结构式相同的, 不同的部分会进行特别说明.
+
+该接口返回通用响应结构, 参考[响应参数](#响应参数--response)
+
+**请求示例：**
+
+```python
+host = "http://127.0.0.1:8888"
+
+def text2img(params: dict) -> dict:
+    """
+    文生图
+    """
+    result = requests.post(url=f"{host}/v1/generation/text-to-image",
+                           data=json.dumps(params),
+                           headers={"Content-Type": "application/json"})
+    return result.json()
+
+result =text2img({
+    "prompt": "1girl sitting on the ground",
+    "async_process": True})
+print(result)
+```
+
+## 图像放大 | image-upscale-vary
+
+该接口对应 Fooocus 中的 Upscale or Variation 功能
+
+该接口参数继承自[文生图](#文生图--text-to-image), 因此后面只会列出和[文生图](#文生图--text-to-image)请求参数差异部分
+
+此外, 该接口提供了两个版本, 两个版本并无功能上的差异, 主要是请求方式略有区别
+
+**基础信息：**
+
+```yaml
+EndPoint_V1: /v1/generation/image-upscale-vary
+EndPoint_V2: /v2/generation/image-upscale-vary
+Method: Post
+DataType: form|json
+```
+
+### V1
+
+**请求参数**
+
+| Name             | Type                | Description                                                                                                               |
+|------------------|---------------------|---------------------------------------------------------------------------------------------------------------------------|
+| input_image      | string($binary)     | 二进制 str 图像                                                                                                                |
+| uov_method       | Enum                | 'Vary (Subtle)','Vary (Strong)','Upscale (1.5x)','Upscale (2x)','Upscale (Fast 2x)','Upscale (Custom)'                    |
+| upscale_value    | float               | 默认为 None , 1.0-5.0, 放大倍数, 仅在 'Upscale (Custom)' 中有效                                                                       |
+| style_selections | List[str]           | 以逗号分割的 Fooocus 风格列表                                                                                                       |
+| loras            | str(List[Lora])     | lora 模型列表, 包含配置, lora 结构: [Lora](#lora), 比如: [{"model_name": "sd_xl_offset_example-lora_1.0.safetensors", "weight": 0.5}] |
+| advanced_params  | str(AdvancedParams) | 高级参数, AdvancedParams 结构 [AdvancedParams](#高级参数--advanceparams), 以字符串形式发送, 可以为空                                            |
+
+**响应参数：**
+
+该接口返回通用响应结构, 参考[响应参数](#响应参数--response)
+
+**请求示例：**
+
+```python
+# 不要加 {"Content-Type": "application/json"} 这个 header
+
+host = "http://127.0.0.1:8888"
+image = open("./examples/imgs/bear.jpg", "rb").read()
+
+def upscale_vary(image, params: dict) -> dict:
+    """
+    Upscale or Vary
+    """
+    response = requests.post(url=f"{host}/v1/generation/image-upscale-vary",
+                        data=params,
+                        files={"input_image": image})
+    return response.json()
+
+result =upscale_vary(image=image,
+                     params={
+                         "uov_method": "Upscale (2x)",
+                         "async_process": True
+                     })
+print(json.dumps(result, indent=4, ensure_ascii=False))
+```
+
+### V2
+
+**请求参数**
+
+| Name          | Type                | Description                                                                                                       |
+|---------------|---------------------|-------------------------------------------------------------------------------------------------------------------|
+| uov_method    | UpscaleOrVaryMethod | 是个枚举类型, 包括 'Vary (Subtle)','Vary (Strong)','Upscale (1.5x)','Upscale (2x)','Upscale (Fast 2x)','Upscale (Custom)' |
+| upscale_value | float               | 默认为 None , 1.0-5.0, 放大倍数, 仅在 'Upscale (Custom)' 中有效                                                               |
+| input_image   | str                 | 输入图像, base64 格式, 或者一个URL                                                                                          |
+
+**响应参数：**
+
+该接口返回通用响应结构, 参考[响应参数](#响应参数--response)
+
+**请求示例：**
+
+```python
+host = "http://127.0.0.1:8888"
+image = open("./examples/imgs/bear.jpg", "rb").read()
+
+def upscale_vary(image, params: dict) -> dict:
+    """
+    Upscale or Vary
+    """
+    params["input_image"] = base64.b64encode(image).decode('utf-8') 
+    response = requests.post(url=f"{host}/v2/generation/image-upscale-vary",
+                        data=json.dumps(params),
+                        headers={"Content-Type": "application/json"},
+                        timeout=300)
+    return response.json()
+
+result =upscale_vary(image=image,
+                     params={
+                         "uov_method": "Upscale (2x)",
+                         "async_process": True
+                     })
+print(json.dumps(result, indent=4, ensure_ascii=False))
+```
+
+## 局部重绘 | image-inpaint-outpaint
+
+**基础信息：**
+
+```yaml
+EndPoint_V1: /v1/generation/image-inpaint-outpaint
+EndPoint_V2: /v2/generation/image-inpaint-outpaint
+Method: Post
+DataType: form|json
+```
+
+### V1
+
+**请求参数**
+
+| Name                      | Type                | Description                                                                                                               |
+|---------------------------|---------------------|---------------------------------------------------------------------------------------------------------------------------|
+| input_image               | string($binary)     | 二进制 str 图像                                                                                                                |
+| input_mask                | string($binary)     | 二进制 str 图像                                                                                                                |
+| inpaint_additional_prompt | string              | 附加描述                                                                                                                      |
+| outpaint_selections       | str                 | 图像扩展方向, 逗号分割的 'Left', 'Right', 'Top', 'Bottom'                                                                            |
+| outpaint_distance_left    | int                 | 图像扩展距离, 默认 0                                                                                                              |
+| outpaint_distance_right   | int                 | 图像扩展距离, 默认 0                                                                                                              |
+| outpaint_distance_top     | int                 | 图像扩展距离, 默认 0                                                                                                              |
+| outpaint_distance_bottom  | int                 | 图像扩展距离, 默认 0                                                                                                              |
+| style_selections          | List[str]           | 以逗号分割的 Fooocus 风格列表                                                                                                       |
+| loras                     | str(List[Lora])     | lora 模型列表, 包含配置, lora 结构: [Lora](#lora), 比如: [{"model_name": "sd_xl_offset_example-lora_1.0.safetensors", "weight": 0.5}] |
+| advanced_params           | str(AdvancedParams) | 高级参数, AdvancedParams 结构 [AdvancedParams](#高级参数--advanceparams), 以字符串形式发送                                                  |
+
+**响应参数：**
+
+该接口返回通用响应结构, 参考[响应参数](#响应参数--response)
+
+**请求示例：**
+
+```python
+# 局部重绘 v1 接口示例
+host = "http://127.0.0.1:8888"
+image = open("./examples/imgs/bear.jpg", "rb").read()
+
+def inpaint_outpaint(params: dict, input_image: bytes, input_mask: bytes = None) -> dict:
+    """
+    局部重绘 v1 接口示例
+    """
+    response = requests.post(url=f"{host}/v1/generation/image-inpaint-outpaint",
+                        data=params,
+                        files={"input_image": input_image,
+                               "input_mask": input_mask})
+    return response.json()
+
+# 图片扩展示例
+result = inpaint_outpaint(params={
+                            "outpaint_selections": "Left,Right",
+                            "async_process": True},
+                          input_image=image,
+                          input_mask=None)
+print(json.dumps(result, indent=4, ensure_ascii=False))
+
+# 局部重绘示例
+source = open("./examples/imgs/s.jpg", "rb").read()
+mask = open("./examples/imgs/m.png", "rb").read()
+result = inpaint_outpaint(params={
+                            "prompt": "a cat",
+                            "async_process": True},
+                          input_image=source,
+                          input_mask=mask)
+print(json.dumps(result, indent=4, ensure_ascii=False))
+```
+
+### V2
+
+**请求参数**
+
+| Name                      | Type                    | Description                                                     |
+|---------------------------|-------------------------|-----------------------------------------------------------------|
+| input_image               | str                     | 输入图像, base64 格式, 或者一个URL                                        |
+| input_mask                | str                     | 输入遮罩, base64 格式, 或者一个URL                                        |
+| inpaint_additional_prompt | str                     | 附加描述词                                                           |
+| outpaint_selections       | List[OutpaintExpansion] | OutpaintExpansion 是一个枚举类型, 值包括 "Left", "Right", "Top", "Bottom" |
+| outpaint_distance_left    | int                     | 图像扩展距离, 默认 0                                                    |
+| outpaint_distance_right   | int                     | 图像扩展距离, 默认 0                                                    |
+| outpaint_distance_top     | int                     | 图像扩展距离, 默认 0                                                    |
+| outpaint_distance_bottom  | int                     | 图像扩展距离, 默认 0                                                    |
+
+**响应参数：**
+
+该接口返回通用响应结构, 参考[响应参数](#响应参数--response)
+
+**请求示例：**
+
+```python
+# 局部重绘 v2 接口示例
+host = "http://127.0.0.1:8888"
+image = open("./examples/imgs/bear.jpg", "rb").read()
+
+def inpaint_outpaint(params: dict) -> dict:
+    """
+    局部重绘 v1 接口示例
+    """
+    response = requests.post(url=f"{host}/v2/generation/image-inpaint-outpaint",
+                        data=json.dumps(params),
+                        headers={"Content-Type": "application/json"})
+    return response.json()
+
+# 图像扩展示例
+result = inpaint_outpaint(params={
+                            "input_image": base64.b64encode(image).decode('utf-8'),
+                            "input_mask": None,
+                            "outpaint_selections": ["Left", "Right"],
+                            "async_process": True})
+print(json.dumps(result, indent=4, ensure_ascii=False))
+
+# 局部重绘示例
+source = open("./examples/imgs/s.jpg", "rb").read()
+mask = open("./examples/imgs/m.png", "rb").read()
+result = inpaint_outpaint(params={
+                            "prompt": "a cat",
+                            "input_image": base64.b64encode(source).decode('utf-8'),
+                            "input_mask": base64.b64encode(mask).decode('utf-8'),
+                            "async_process": True})
+print(json.dumps(result, indent=4, ensure_ascii=False))
+```
+
+## 图生图 | image-prompt
+
+该接口更新自 `v0.3.27` 后有重大更新。从继承自 [文生图](#文生图--text-to-image) 更改为继承自 [局部重绘](#局部重绘--image-inpaint-outpaint)
+
+该版本之后可以通过该接口实现 `inpaint_outpaint` 以及 `image-prompt` 接口的功能
+
+> 多功能接口，并非可以同时实现 `inpaint_outpaint` 以及 `image-prompt` 接口的功能
+
+**基础信息：**
+
+```yaml
+EndPoint_V1: /v1/generation/image-prompt
+EndPoint_V2: /v2/generation/image-prompt
+Method: Post
+DataType: form|json
+```
+
+### V1
+
+**请求参数**
+
+> 注意: 虽然接口更改为继承自[局部重绘](#局部重绘--image-inpaint-outpaint), 但下方表格展示的仍然继承自[文生图](#文生图--text-to-image), 但参数是完整的
+
+| Name                      | Type                | Description                                                                                                               |
+|---------------------------|---------------------|---------------------------------------------------------------------------------------------------------------------------|
+| input_image               | Bytes               | 二进制图像, 用于局部重绘                                                                                                             |
+| input_mask                | Bytes               | 二进制图像遮罩, 用于局部重绘                                                                                                           |
+| inpaint_additional_prompt | str                 | inpaint 附加提示词                                                                                                             |
+| outpaint_selections       | str                 | 图像扩展选项, 逗号分割的 "Left", "Right", "Top", "Bottom"                                                                            |
+| outpaint_distance_left    | int                 | 图像扩展距离, 默认 0                                                                                                              |
+| outpaint_distance_right   | int                 | 图像扩展距离, 默认 0                                                                                                              |
+| outpaint_distance_top     | int                 | 图像扩展距离, 默认 0                                                                                                              |
+| outpaint_distance_bottom  | int                 | 图像扩展距离, 默认 0                                                                                                              |
+| cn_img1                   | string($binary)     | 二进制 str 图像                                                                                                                |
+| cn_stop1                  | float               | 默认 0.6                                                                                                                    |
+| cn_weight1                | float               | 默认 0.6                                                                                                                    |
+| cn_type1                  | Enum                | "ImagePrompt", "FaceSwap", "PyraCanny", "CPDS" 中的一个                                                                       |
+| cn_img2                   | string($binary)     | 二进制 str 图像                                                                                                                |
+| cn_stop2                  | float               | 默认 0.6                                                                                                                    |
+| cn_weight2                | float               | 默认 0.6                                                                                                                    |
+| cn_type2                  | Enum                | "ImagePrompt", "FaceSwap", "PyraCanny", "CPDS" 中的一个                                                                       |
+| cn_img3                   | string($binary)     | 二进制 str 图像                                                                                                                |
+| cn_stop3                  | float               | 默认 0.6                                                                                                                    |
+| cn_weight3                | float               | 默认 0.6                                                                                                                    |
+| cn_type3                  | Enum                | "ImagePrompt", "FaceSwap", "PyraCanny", "CPDS" 中的一个                                                                       |
+| cn_img4                   | string($binary)     | 二进制 str 图像                                                                                                                |
+| cn_stop4                  | float               | 默认 0.6                                                                                                                    |
+| cn_weight4                | float               | 默认 0.6                                                                                                                    |
+| cn_type4                  | Enum                | "ImagePrompt", "FaceSwap", "PyraCanny", "CPDS" 中的一个                                                                       |
+| style_selections          | List[str]           | 以逗号分割的 Fooocus 风格列表                                                                                                       |
+| loras                     | str(List[Lora])     | lora 模型列表, 包含配置, lora 结构: [Lora](#lora), 比如: [{"model_name": "sd_xl_offset_example-lora_1.0.safetensors", "weight": 0.5}] |
+| advanced_params           | str(AdvancedParams) | 高级参数, AdvancedParams 结构 [AdvancedParams](#高级参数--advanceparams), 以字符串形式发送                                                  |
+
+**响应参数：**
+
+该接口返回通用响应结构, 参考[响应参数](#响应参数--response)
+
+**请求示例：**
+
+```python
+# image_prompt v1 接口示例
+host = "http://127.0.0.1:8888"
+image = open("./examples/imgs/bear.jpg", "rb").read()
+source = open("./examples/imgs/s.jpg", "rb").read()
+mask = open("./examples/imgs/m.png", "rb").read()
+
+def image_prompt(params: dict,
+                 input_image: bytes=None,
+                 input_mask: bytes=None,
+                 cn_img1: bytes=None,
+                 cn_img2: bytes=None,
+                 cn_img3: bytes=None,
+                 cn_img4: bytes=None,) -> dict:
+    """
+    image prompt
+    """
+    response = requests.post(url=f"{host}/v1/generation/image-prompt",
+                             data=params,
+                             files={
+                                 "input_image": input_image,
+                                 "input_mask": input_mask,
+                                 "cn_img1": cn_img1,
+                                 "cn_img2": cn_img2,
+                                 "cn_img3": cn_img3,
+                                 "cn_img4": cn_img4,
+                              })
+    return response.json()
+
+# 图像扩展
+params = {
+    "outpaint_selections": ["Left", "Right"],
+    "image_prompts": [] # 必传参数，可以为空列表
+}
+result = image_prompt(params=params, input_iamge=image)
+print(json.dumps(result, indent=4, ensure_ascii=False))
+
+# 局部重绘
+
+params = {
+    "prompt": "1girl sitting on the chair",
+    "image_prompts": [], # 必传参数，可以为空列表
+    "async_process": True
+}
+result = image_prompt(params=params, input_iamge=source, input_mask=mask)
+print(json.dumps(result, indent=4, ensure_ascii=False))
+
+# image prompt
+
+params = {
+    "prompt": "1girl sitting on the chair",
+    "image_prompts": [
+        {
+            "cn_stop": 0.6,
+            "cn_weight": 0.6,
+            "cn_type": "ImagePrompt"
+        },{
+            "cn_stop": 0.6,
+            "cn_weight": 0.6,
+            "cn_type": "ImagePrompt"
+        }]
+    }
+result = image_prompt(params=params, cn_img1=image, cn_img2=source)
+print(json.dumps(result, indent=4, ensure_ascii=False))
+```
+
+### V2
+
+**请求参数**
+
+| Name                      | Type                    | Description                                    |
+|---------------------------|-------------------------|------------------------------------------------|
+| input_image               | str                     | base64 图像, 或者一个URL, 用于局部重绘                     |
+| input_mask                | str                     | base64 图像遮罩, 或者一个URL, 用于局部重绘                   |
+| inpaint_additional_prompt | str                     | inpaint 附加提示词                                  |
+| outpaint_selections       | List[OutpaintExpansion] | 图像扩展选项, 逗号分割的 "Left", "Right", "Top", "Bottom" |
+| outpaint_distance_left    | int                     | 图像扩展距离, 默认 0                                   |
+| outpaint_distance_right   | int                     | 图像扩展距离, 默认 0                                   |
+| outpaint_distance_top     | int                     | 图像扩展距离, 默认 0                                   |
+| outpaint_distance_bottom  | int                     | 图像扩展距离, 默认 0                                   |
+| image_prompts             | List[ImagePrompt]       | 图像列表, 包含配置, ImagePrompt 结构如下：                  |
+
+**ImagePrompt**
+
+| Name      | Type           | Description                                                         |
+|-----------|----------------|---------------------------------------------------------------------|
+| cn_img    | str            | 输入图像, base64 编码, 或者一个URL                                            |
+| cn_stop   | float          | 停止位置, 范围 0-1, 默认 0.5                                                |
+| cn_weight | float          | 权重, 范围 0-2, 默认 1.0                                                  |
+| cn_type   | ControlNetType | 控制网络类型, 是一个枚举类型, 包括: "ImagePrompt", "FaceSwap", "PyraCanny", "CPDS" |
+
+**响应参数：**
+
+该接口返回通用响应结构, 参考[响应参数](#响应参数--response)
+
+**请求示例：**
+
+```python
+# image_prompt v2 接口示例
+host = "http://127.0.0.1:8888"
+image = open("./examples/imgs/bear.jpg", "rb").read()
+source = open("./examples/imgs/s.jpg", "rb").read()
+mask = open("./examples/imgs/m.png", "rb").read()
+
+def image_prompt(params: dict) -> dict:
+    """
+    image prompt
+    """
+    response = requests.post(url=f"{host}/v2/generation/image-prompt",
+                             data=json.dumps(params),
+                             headers={"Content-Type": "application/json"})
+    return response.json()
+
+# 图像扩展
+params = {
+    "input_image": base64.b64encode(image).decode('utf-8'),
+    "outpaint_selections": ["Left", "Right"],
+    "image_prompts": [] # 必传参数，可以为空列表
+}
+result = image_prompt(params)
+print(json.dumps(result, indent=4, ensure_ascii=False))
+
+# 局部重绘
+
+params = {
+    "prompt": "1girl sitting on the chair",
+    "input_image": base64.b64encode(source).decode('utf-8'),
+    "input_mask": base64.b64encode(mask).decode('utf-8'),
+    "image_prompts": [], # 必传参数，可以为空列表
+    "async_process": True
+}
+result = image_prompt(params)
+print(json.dumps(result, indent=4, ensure_ascii=False))
+
+# image prompt
+
+params = {
+    "prompt": "1girl sitting on the chair",
+    "image_prompts": [
+        {
+            "cn_img": base64.b64encode(source).decode('utf-8'),
+            "cn_stop": 0.6,
+            "cn_weight": 0.6,
+            "cn_type": "ImagePrompt"
+        },{
+            "cn_img": base64.b64encode(image).decode('utf-8'),
+            "cn_stop": 0.6,
+            "cn_weight": 0.6,
+            "cn_type": "ImagePrompt"
+        }]
+    }
+result = image_prompt(params)
+print(json.dumps(result, indent=4, ensure_ascii=False))
+```
+
+## text to image with image prompt
+
+该接口暂无 v1 版本
+
+**基础信息：**
+
+```yaml
+EndPoint: /v2/generation/text-to-image-with-ip
+Method: Post
+DataType: json
+```
+
+**请求参数**
+
+| Name          | Type              | Description |
+|---------------|-------------------|-------------|
+| image_prompts | List[ImagePrompt] | 图像列表        |
+
+**请求示例**:
+
+```python
+# text to image with image prompt 示例
+host = "http://127.0.0.1:8888"
+image = open("./examples/imgs/bear.jpg", "rb").read()
+source = open("./examples/imgs/s.jpg", "rb").read()
+def image_prompt(params: dict) -> dict:
+    """
+    image prompt
+    """
+    response = requests.post(url=f"{host}/v2/generation/text-to-image-with-ip",
+                             data=json.dumps(params),
+                             headers={"Content-Type": "application/json"})
+    return response.json()
+
+params = {
+    "prompt": "A bear",
+    "image_prompts": [
+        {
+            "cn_img": base64.b64encode(source).decode('utf-8'),
+            "cn_stop": 0.6,
+            "cn_weight": 0.6,
+            "cn_type": "ImagePrompt"
+        },{
+            "cn_img": base64.b64encode(image).decode('utf-8'),
+            "cn_stop": 0.6,
+            "cn_weight": 0.6,
+            "cn_type": "ImagePrompt"
+        }
+    ]
+}
+result = image_prompt(params)
+print(json.dumps(result, indent=4, ensure_ascii=False))
+```
+
+## 图像反推 | describe
+
+**基础信息：**
+
+```yaml
+EndPoint: /v1/tools/describe-image
+Method: Post
+DataType: form
+```
+
+**请求参数**
+
+| Name | Type | Description                 |
+|------|------|-----------------------------|
+| type | Enum | 反推类型, "Photo", "Anime" 中的一个 |
+
+**请求示例**:
+
+```python
+def describe_image(image: bytes,
+                   params: dict = {"type": "Photo"}) -> dict:
+    """
+    describe-image
+    """
+    response = requests.post(url="http://127.0.0.1:8888/v1/tools/describe-image",
+                        params=params,
+                        files={
+                            "image": image
+                        },
+                        timeout=30)
+    return response.json()
+```
+
+**响应示例**:
+
+```python
+{
+  "describe": "a young woman posing with her hands behind her head"
+}
+```
+
+--------------------------------------------
+
+## 列出模型 | all-models
+
+**基础信息：**
+
+```yaml
+EndPoint: /v1/engines/all-models
+Method: Get
+```
+
+**请求示例**:
+
+```python
+def all_models() -> dict:
+    """
+    all-models
+    """
+    response = requests.get(url="http://127.0.0.1:8888/v1/engines/all-models",
+                        timeout=30)
+    return response.json()
+```
+
+**响应示例**:
+
+```python
+{
+  "model_filenames": [
+    "juggernautXL_version6Rundiffusion.safetensors",
+    "sd_xl_base_1.0_0.9vae.safetensors",
+    "sd_xl_refiner_1.0_0.9vae.safetensors"
+  ],
+  "lora_filenames": [
+    "sd_xl_offset_example-lora_1.0.safetensors"
+  ]
+}
+```
+
+## 刷新模型 | refresh-models
+
+**基础信息：**
+
+> 该接口已移除，功能合并到 [列出模型](#列出模型--all-models)
+
+```yaml
+EndPoint: /v1/engines/refresh-models
+Method: Post
+```
+
+**请求示例**
+```python
+def refresh() -> dict:
+    """
+    refresh-models
+    """
+    response = requests.post(url="http://127.0.0.1:8888/v1/engines/refresh-models",
+                        timeout=30)
+    return response.json()
+```
+
+**响应示例**
+```python
+{
+  "model_filenames": [
+    "juggernautXL_version6Rundiffusion.safetensors",
+    "sd_xl_base_1.0_0.9vae.safetensors",
+    "sd_xl_refiner_1.0_0.9vae.safetensors"
+  ],
+  "lora_filenames": [
+    "sd_xl_offset_example-lora_1.0.safetensors"
+  ]
+}
+```
+
+## 样式 | styles
+
+**基础信息：**
+
+```yaml
+EndPoint: /v1/engines/styles
+Method: Get
+```
+
+**请求示例**:
+
+```python
+def styles() -> dict:
+    """
+    styles
+    """
+    response = requests.get(url="http://127.0.0.1:8888/v1/engines/styles",
+                        timeout=30)
+    return response.json()
+```
+
+**响应示例**:
+
+```python
+[
+  "Fooocus V2",
+  "Fooocus Enhance",
+  ...
+  "Watercolor 2",
+  "Whimsical And Playful"
+]
+```
+
+# Fooocus API 任务相关接口
+
+## 任务队列 | job-queue
+
+**基础信息：**
+
+```yaml
+EndPoint: /v1/engines/job-queue
+Method: Get
+```
+
+**请求示例**:
+
+```python
+def job_queue() -> dict:
+    """
+    job-queue
+    """
+    response = requests.get(url="http://127.0.0.1:8888/v1/generation/job-queue",
+                        timeout=30)
+    return response.json()
+```
+
+**响应示例**:
+
+```python
+{
+  "running_size": 0,
+  "finished_size": 1,
+  "last_job_id": "cac3914a-926d-4b6f-a46a-83794a0ce1d4"
+}
+```
+
+## 查询任务 | query-job
+
+**基础信息：**
+
+```yaml
+EndPoint: /v1/generation/query-job
+Method: Get
+```
+
+**请求示例**:
+```python
+def taskResult(task_id: str) -> dict:
+    # 获取任务状态
+    task_status = requests.get(url="http://127.0.0.1:8888/v1/generation/query-job",
+                               params={"job_id": task_id,
+                                       "require_step_preview": False},
+                               timeout=30)
+
+    return task_status.json()
+```
+
+**响应示例**:
+```python
+{
+  "job_id": "cac3914a-926d-4b6f-a46a-83794a0ce1d4",
+  "job_type": "Text to Image",
+  "job_stage": "SUCCESS",
+  "job_progress": 100,
+  "job_status": "Finished",
+  "job_step_preview": null,
+  "job_result": [
+    {
+      "base64": null,
+      "url": "http://127.0.0.1:8888/files/2023-11-27/b928e50e-3c09-4187-a3f9-1c12280bfd95.png",
+      "seed": 8228839561385006000,
+      "finish_reason": "SUCCESS"
+    }
+  ]
+}
+```
+
+## 查询任务历史 | job-history
+
+**基础信息：**
+
+```yaml
+EndPoint: /v1/generation/job-history
+Method: get
+```
+
+**请求示例**:
+
+```python
+def job-history() -> dict:
+    """
+    job-history
+    """
+    response = requests.get(url="http://127.0.0.1:8888/v1/generation/job-history",
+                        timeout=30)
+    return response.json()
+```
+
+**响应示例**:
+
+```python
+{
+  "queue": [],
+  "history": [
+    "job_id": "cac3914a-926d-4b6f-a46a-83794a0ce1d4",
+    "is_finished": True
+  ]
+}
+```
+
+## 停止任务 | stop
+
+**基础信息：**
+
+```yaml
+EndPoint: /v1/generation/stop
+Method: post
+```
+
+**请求示例**:
+
+```python
+def stop() -> dict:
+    """
+    stop
+    """
+    response = requests.post(url="http://127.0.0.1:8888/v1/generation/stop",
+                        timeout=30)
+    return response.json()
+```
+
+**响应示例**:
+
+```python
+{
+  "msg": "success"
+}
+```
+
+## ping
+
+**基础信息：**
+
+```yaml
+EndPoint: /ping
+Method: get
+```
+
+pong
+
+# webhook
+
+你可以在命令行通过 `--webhook-url` 指定一个地址，以便异步任务完成之后可以收到通知
+
+下面是一个简单的示例来展示 `webhook` 是如何工作的
+
+首先，使用下面的代码启动一个简易服务器:
+
+```python
+from fastapi import FastAPI
+import uvicorn
+
+app = FastAPI()
+
+@app.post("/status")
+async def status(requests: dict):
+    print(requests)
+
+uvicorn.run(app, host="0.0.0.0", port=8000)
+```
+
+然后, 在启动 Fooocus API 时添加 `--webhook-url http://host:8000/status`
+
+通过任意方式提交一个任务, 等完成后你会在这个简易服务器的后台看到任务结束信息：
+
+```python
+{'job_id': '717ec0b5-85df-4174-80d6-bddf93cd8248', 'job_result': [{'url': 'http://127.0.0.1:8888/files/2023-12-29/f1eca704-718e-4781-9d5f-82d41aa799d7.png', 'seed': '3283449865282320931'}]}
+```
+
+# 公共请求体
+
+## 高级参数 | AdvanceParams
+
+| Name                                 | Type  | Description                                                           |
+|--------------------------------------|-------|-----------------------------------------------------------------------|
+| disable_preview                      | bool  | 是否禁用预览, 默认 False                                                      |
+| disable_intermediate_results         | bool  | 是否禁用中间结果, 默认 False                                                    |
+| disable_seed_increment               | bool  | 是否禁用种子递增, 默认 False                                                    |
+| adm_scaler_positive                  | float | 正 ADM Guidance Scaler, 默认 1.5, 范围 0.1-3.0                             |
+| adm_scaler_negative                  | float | 负 ADM Guidance Scaler, 默认 0.8, 范围 0.1-3.0                             |
+| adm_scaler_end                       | float | ADM Guidance Scaler 结束值, 默认 0.5, 范围 0.0-1.0                           |
+| refiner_swap_method                  | str   | 优化模型交换方法, 默认 `joint`                                                  |
+| adaptive_cfg                         | float | CFG Mimicking from TSNR, 默认 7.0, 范围 1.0-30.0                          |
+| sampler_name                         | str   | 采样器, 默认 `default_sampler`                                             |
+| scheduler_name                       | str   | 调度器, 默认 `default_scheduler`                                           |
+| overwrite_step                       | int   | Forced Overwrite of Sampling Step, 默认 -1, 范围 -1-200                   |
+| overwrite_switch                     | int   | Forced Overwrite of Refiner Switch Step, 默认 -1, 范围 -1-200             |
+| overwrite_width                      | int   | Forced Overwrite of Generating Width, 默认 -1, 范围 -1-2048               |
+| overwrite_height                     | int   | Forced Overwrite of Generating Height, 默认 -1, 范围 -1-2048              |
+| overwrite_vary_strength              | float | Forced Overwrite of Denoising Strength of "Vary", 默认 -1, 范围 -1-1.0    |
+| overwrite_upscale_strength           | float | Forced Overwrite of Denoising Strength of "Upscale", 默认 -1, 范围 -1-1.0 |
+| mixing_image_prompt_and_vary_upscale | bool  | Mixing Image Prompt and Vary/Upscale, 默认 False                        |
+| mixing_image_prompt_and_inpaint      | bool  | Mixing Image Prompt and Inpaint, 默认 False                             |
+| debugging_cn_preprocessor            | bool  | Debug Preprocessors, 默认 False                                         |
+| skipping_cn_preprocessor             | bool  | Skip Preprocessors, 默认 False                                          |
+| controlnet_softness                  | float | Softness of ControlNet, 默认 0.25, 范围 0.0-1.0                           |
+| canny_low_threshold                  | int   | Canny Low Threshold, 默认 64, 范围 1-255                                  |
+| canny_high_threshold                 | int   | Canny High Threshold, 默认 128, 范围 1-255                                |
+| freeu_enabled                        | bool  | FreeU enabled, 默认 False                                               |
+| freeu_b1                             | float | FreeU B1, 默认 1.01                                                     |
+| freeu_b2                             | float | FreeU B2, 默认 1.02                                                     |
+| freeu_s1                             | float | FreeU B3, 默认 0.99                                                     |
+| freeu_s2                             | float | FreeU B4, 默认 0.95                                                     |
+| debugging_inpaint_preprocessor       | bool  | Debug Inpaint Preprocessing, 默认 False                                 |
+| inpaint_disable_initial_latent       | bool  | Disable initial latent in inpaint, 默认 False                           |
+| inpaint_engine                       | str   | Inpaint Engine, 默认 `v2.6`                                             |
+| inpaint_strength                     | float | Inpaint Denoising Strength, 默认 1.0, 范围 0.0-1.0                        |
+| inpaint_respective_field             | float | Inpaint Respective Field, 默认 1.0, 范围 0.0-1.0                          |
+| inpaint_mask_upload_checkbox         | bool  | 是否上传掩码图片, 默认 False                                                    |
+| invert_mask_checkbox                 | bool  | 是否反转掩码, 默认 False                                                      |
+| inpaint_erode_or_dilate              | int   | Mask Erode or Dilate, 默认 0, 范围 -64-64                                 |
+
+## lora
+
+| Name       | Type  | Description |
+|------------|-------|-------------|
+| enabled    | bool  | 是否启用lora    |
+| model_name | str   | 模型名称        |
+| weight     | float | 权重, 默认 0.5  |
+
+## 响应参数 | response
+
+成功响应：
+
+**async_process: True**
+
+| Name             | Type  | Description |
+|------------------|-------|-------------|
+| job_id           | int   | 任务ID        |
+| job_type         | str   | 任务类型        |
+| job_stage        | str   | 任务阶段        |
+| job_progress     | float | 任务进度        |
+| job_status       | str   | 任务状态        |
+| job_step_preview | str   | 任务预览        |
+| job_result       | str   | 任务结果        |
+
+**async_process: False**
+
+| Name          | Type | Description                                  |
+|---------------|------|----------------------------------------------|
+| base64        | str  | 图片base64编码, 根据 `require_base64` 参数决定是否为 null |
+| url           | str  | 图片url                                        |
+| seed          | int  | 图片种子                                         |
+| finish_reason | str  | 任务结束原因                                       |
+
+失败响应：

docs/change_logs.md

@@ -0,0 +1,178 @@
+# ChangeLog for Fooocus-API
+
+## [v0.5.0.1]
+### Changed
+- Fooocus to v2.5.3
+- Add enhance image endpoint
+- Add generate mask endpoint
+- Influenced by Fooocus, the worker.py was reconstructed
+- Update docs
+- Returned base64 str now include identifier like this `data:image/jpeg;base64,`
+
+### Fixed
+- Issue #375
+- Issue #378
+- Save extension params now also takes effect for base64 str in the returned data
+
+## [v0.4.1.1] - 2024-05-2
+### Added
+- LoraManager support tar file
+
+### Changed
+- Update Fooocus to v2.4.3
+
+### Fixed
+- Issue #363
+
+## [v0.4.1.0] - 2024-05-28
+### Added
+- Add nsfw checker
+
+### Changed
+- Fooocus to v2.4.1
+- Increase support for config.txt
+
+### Fixed
+- Issue #319
+- Issue #327
+- Issue #332
+- Fix JPEG support
+
+## [v0.4.0.6] - 2024-04-24
+### Added
+- UA in image request
+
+### Changed
+- Fooocus to v2.3.1
+- Code formatting.
+
+### Fixed
+- Issue #302
+- Issue #294
+
+## [v0.4.0.5] - 2024-04-16
+### Changed
+- Sync v1 endpoints.
+
+### Fixed
+- meta_scheme error
+
+## [v0.4.0.4] - 2024-04-15
+### Added
+- Image meta save to image
+
+### Changed
+- Code formatting.
+
+## [v0.4.0.3] - 2024-04-12
+### Fixed
+- Fix opencv-python-headless failed in Docker
+- Issue #270
+
+## [v0.4.0.2] - 2024-04-09
+### Fixed
+- Issue #280
+- Issue #222
+
+## [v0.4.0.1] - 2024-04-08
+### Added
+- Url support for lora in replicate
+
+## [v0.4.0.0] - 2024-04-08
+### Changed
+- Update docs
+- Rewrite Dockerfile
+- Rewirte examples
+- Full code of Fooocus include
+- Remove related code
+- Optimize project structure.
+
+## [v0.3.33] - 2024-04-07
+### Added
+- Support for Lightning model.
+- Update default checkpoint for Replicate.
+
+### Fixed
+- Issue #244
+- Issue #259
+- Issue #232
+
+## [v0.3.32] - 2024-03-21
+### Added
+- Support for Fooocus 2.3.0.
+
+### Changed
+- Project structure optimization.
+
+### Fixed
+- Removed unnecessary dependencies and optimized code.
+
+## [v0.3.31] - 2024-03-20
+### Added
+- Save extension
+- Secure API future with the use of API keys.
+- Add seeds for predict output
+
+### Changed
+- Update Docs
+
+### Fixed
+- OOM when running a long time
+- Some spell error
+- Issue with the Fooocus-API version in startup output.
+
+## [v0.3.30] - 2024-01-26
+### Added
+- Support url for input_image in v2 API.
+- Image Prompt Mixing requirements implemented
+- Add SQLite database support for history.
+
+### Changed
+- Update Docs
+- Large queue size support
+- Optimized async task response when the queue is full
+- Update cog branch
+- Optimized cli flages parser.
+- Optimized some code formatting.
+- Optimized the underlying logic of task execution.
+- Default queue history size to 0 for no limit.
+
+### Fixed
+- Fix condition. default params broke here and this fixes auto mixing feature.
+- Fix error when use `Extreme Speed' with cog.
+- Fix typo of 'presistent'
+- Image Prompt Mixing requirements implemented
+- Some spell error, some translations.
+- Fix image prompt must require 'input_image'.
+- Implemented support for storing history to the database.
+
+## [v0.3.29] - 2024-01-04
+### Added
+- Add example using ipynb
+- Add error logging
+- Add check for aspect_ratios_selection
+- Image Prompt Mixing requirements implemented.
+
+### Changed
+- Update Docs
+- Merge Fooocus to v2.1.860
+
+### Fixed
+- Various bugs and issues reported by the community.
+
+## [v0.3.28] - 2024-01-03
+### Added
+- Add ping endpoint
+- Describe interface to get prompts from images.
+- Add image_prompt to text2img endpoint
+- Add mirror for fooocus
+
+### Changed
+- Update Docs
+- Added query job history API and webhook_url support for each generation request.
+- Change to exit when Fooocus check failed
+
+### Fixed
+- Fix #122 query job not found error.
+
+Please note that this ChangeLog is a summary and may not include all changes. For a complete list of changes, please refer to the commit history on the [Fooocus-API GitHub repository](https://github.com/mrhan1993/Fooocus-API).

docs/change_logs_zh.md

@@ -0,0 +1,178 @@
+# ChangeLog for Fooocus-API
+
+## [v0.5.0.1]
+### Changed
+- 合并到 Fooocus 2.5.3
+- 增加图像增强接口
+- 增加遮罩生成接口
+- 受 Fooocus 影响，重构了 worker.py
+- 更新文档
+- 返回数据中的 base64 字符串现在包含图像标识符，比如 `data:image/jpeg;base64,`
+
+### Fixed
+- Issue #375
+- Issue #378
+- 保存扩展名参数现在可以对返回的 base64 生效了
+
+## [v0.4.1.1] - 2024-05-2
+### Added
+- LoraManager 支持 tar 格式压缩文件
+
+### Changed
+- Update Fooocus to v2.4.3
+
+### Fixed
+- Issue #363
+
+## [v0.4.1.0] - 2024-05-28
+### Added
+- 黄图检测
+
+### Changed
+- Fooocus to v2.4.1
+- 增强对 config.txt 文件的支持
+
+### Fixed
+- Issue #319
+- Issue #327
+- Issue #332
+- Fix JPEG support
+
+## [v0.4.0.6] - 2024-04-24
+### Added
+- UA in image request
+
+### Changed
+- Fooocus to v2.3.1
+- Code formatting.
+
+### Fixed
+- Issue #302
+- Issue #294
+
+## [v0.4.0.5] - 2024-04-16
+### Changed
+- Sync v1 endpoints.
+
+### Fixed
+- meta_scheme error
+
+## [v0.4.0.4] - 2024-04-15
+### Added
+- Image meta save to image
+
+### Changed
+- Code formatting.
+
+## [v0.4.0.3] - 2024-04-12
+### Fixed
+- Fix opencv-python-headless failed in Docker
+- Issue #270
+
+## [v0.4.0.2] - 2024-04-09
+### Fixed
+- Issue #280
+- Issue #222
+
+## [v0.4.0.1] - 2024-04-08
+### Added
+- Url support for lora in replicate
+
+## [v0.4.0.0] - 2024-04-08
+### Changed
+- Update docs
+- Rewrite Dockerfile
+- Rewirte examples
+- Full code of Fooocus include
+- Remove related code
+- Optimize project structure.
+
+## [v0.3.33] - 2024-04-07
+### Added
+- Support for Lightning model.
+- Update default checkpoint for Replicate.
+
+### Fixed
+- Issue #244
+- Issue #259
+- Issue #232
+
+## [v0.3.32] - 2024-03-21
+### Added
+- Support for Fooocus 2.3.0.
+
+### Changed
+- Project structure optimization.
+
+### Fixed
+- Removed unnecessary dependencies and optimized code.
+
+## [v0.3.31] - 2024-03-20
+### Added
+- Save extension
+- Secure API future with the use of API keys.
+- Add seeds for predict output
+
+### Changed
+- Update Docs
+
+### Fixed
+- OOM when running a long time
+- Some spell error
+- Issue with the Fooocus-API version in startup output.
+
+## [v0.3.30] - 2024-01-26
+### Added
+- Support url for input_image in v2 API.
+- Image Prompt Mixing requirements implemented
+- Add SQLite database support for history.
+
+### Changed
+- Update Docs
+- Large queue size support
+- Optimized async task response when the queue is full
+- Update cog branch
+- Optimized cli flages parser.
+- Optimized some code formatting.
+- Optimized the underlying logic of task execution.
+- Default queue history size to 0 for no limit.
+
+### Fixed
+- Fix condition. default params broke here and this fixes auto mixing feature.
+- Fix error when use `Extreme Speed' with cog.
+- Fix typo of 'presistent'
+- Image Prompt Mixing requirements implemented
+- Some spell error, some translations.
+- Fix image prompt must require 'input_image'.
+- Implemented support for storing history to the database.
+
+## [v0.3.29] - 2024-01-04
+### Added
+- Add example using ipynb
+- Add error logging
+- Add check for aspect_ratios_selection
+- Image Prompt Mixing requirements implemented.
+
+### Changed
+- Update Docs
+- Merge Fooocus to v2.1.860
+
+### Fixed
+- Various bugs and issues reported by the community.
+
+## [v0.3.28] - 2024-01-03
+### Added
+- Add ping endpoint
+- Describe interface to get prompts from images.
+- Add image_prompt to text2img endpoint
+- Add mirror for fooocus
+
+### Changed
+- Update Docs
+- Added query job history API and webhook_url support for each generation request.
+- Change to exit when Fooocus check failed
+
+### Fixed
+- Fix #122 query job not found error.
+
+Please note that this ChangeLog is a summary and may not include all changes. For a complete list of changes, please refer to the commit history on the [Fooocus-API GitHub repository](https://github.com/mrhan1993/Fooocus-API).

docs/migrate.md

@@ -0,0 +1,377 @@
+# 参数对照表
+
+In next table, `AdvancedParams` is replaced with `adp`, and the rule of name change is to unify with Fooocus.
+
+| Fooocus-API                              | FooocusAPI                           | 备注                  |
+|------------------------------------------|--------------------------------------|---------------------|
+| prompt                                   | prompt                               |                     |
+| negative_prompt                          | negative_prompt                      |                     |
+| style_selections                         | style_selections                     |                     |
+| performance_selection                    | performance_selection                |                     |
+| aspect_ratios_selection                  | aspect_ratios_selection              |                     |
+| image_number                             | image_number                         |                     |
+| image_seed                               | image_seed                           |                     |
+| sharpness                                | sharpness                            |                     |
+| guidance_scale                           | guidance_scale                       |                     |
+| base_model_name                          | base_model_name                      |                     |
+| refiner_model_name                       | refiner_model_name                   |                     |
+| refiner_switch                           | refiner_switch                       |                     |
+| loras                                    | loras                                | same, a list of lora obj |
+|                                          | input_image_checkbox                 | this is always true      |
+|                                          | current_tab                          | need not care this      |
+| uov_method                               | uov_method                           |                     |
+| **input_image**                          | **uov_input_image**                  | use variable name in Fooocus    |
+| outpaint_selections                      | outpaint_selections                  |                     |
+| **input_image**                          | **inpaint_input_image**              | use variable name in Fooocus    |
+| inpaint_additional_prompt                | inpaint_additional_prompt            |                     |
+| **input_mask**                           | **inpaint_mask_image_upload**        | use variable name in Fooocus    |
+| adp.disable_preview                      | disable_preview                      |                     |
+| adp.disable_intermediate_results         | disable_intermediate_results         |                     |
+| adp.disable_seed_increment               | disable_seed_increment               |                     |
+| adp.black_out_nsfw                       | black_out_nsfw                       |                     |
+| adp.adm_scaler_positive                  | adm_scaler_positive                  |                     |
+| adp.adm_scaler_negative                  | adm_scaler_negative                  |                     |
+| adp.adm_scaler_end                       | adm_scaler_end                       |                     |
+| adp.adaptive_cfg                         | adaptive_cfg                         |                     |
+| adp.clip_skip                            | clip_skip                            |                     |
+| adp.sampler_name                         | sampler_name                         |                     |
+| adp.scheduler_name                       | scheduler_name                       |                     |
+| adp.vae_name                             | vae_name                             |                     |
+| adp.overwrite_step                       | overwrite_step                       |                     |
+| adp.overwrite_switch                     | overwrite_switch                     |                     |
+| adp.overwrite_width                      | overwrite_width                      |                     |
+| adp.overwrite_height                     | overwrite_height                     |                     |
+| adp.overwrite_vary_strength              | overwrite_vary_strength              |                     |
+| adp.overwrite_upscale_strength           | overwrite_upscale_strength           |                     |
+| adp.mixing_image_prompt_and_vary_upscale | mixing_image_prompt_and_vary_upscale |                     |
+| adp.mixing_image_prompt_and_inpaint      | mixing_image_prompt_and_inpaint      |                     |
+| adp.debugging_cn_preprocessor            | debugging_cn_preprocessor            |                     |
+| adp.skipping_cn_preprocessor             | skipping_cn_preprocessor             |                     |
+| adp.canny_low_threshold                  | canny_low_threshold                  |                     |
+| adp.canny_high_threshold                 | canny_high_threshold                 |                     |
+| adp.refiner_swap_method                  | refiner_swap_method                  |                     |
+| adp.controlnet_softness                  | controlnet_softness                  |                     |
+| adp.freeu_enabled                        | freeu_enabled                        |                     |
+| adp.freeu_b1                             | freeu_b1                             |                     |
+| adp.freeu_b2                             | freeu_b2                             |                     |
+| adp.freeu_s1                             | freeu_s1                             |                     |
+| adp.freeu_s2                             | freeu_s2                             |                     |
+| adp.debugging_inpaint_preprocessor       | debugging_inpaint_preprocessor       |                     |
+| adp.inpaint_disable_initial_latent       | inpaint_disable_initial_latent       |                     |
+| adp.inpaint_engine                       | inpaint_engine                       |                     |
+| adp.inpaint_strength                     | inpaint_strength                     |                     |
+| adp.inpaint_respective_field             | inpaint_respective_field             |                     |
+| adp.inpaint_mask_upload_checkbox         | inpaint_mask_upload_checkbox         |                     |
+| adp.invert_mask_checkbox                 | invert_mask_checkbox                 |                     |
+| adp.inpaint_erode_or_dilate              | inpaint_erode_or_dilate              |                     |
+| **image_prompts**                        | **controlnet_image**                 | just change name            |
+|                                          | generate_image_grid                  | new, default is better     |
+| outpaint_distance_left                   | outpaint_distance                    | merge these to one       |
+| outpaint_distance_right                  |                                      | use a list to pass these four      |
+| outpaint_distance_top                    |                                      | exp: [100, 50, 0, 0]  |
+| outpaint_distance_bottom                 |                                      | Directions are: left, up, right, down      |
+| **upscale_value**                        | **upscale_multiple**                 | name change only               |
+|                                          | preset                               | new, use this this specified preset  |
+|                                          | stream_output                        | new, similar to LLM streaming output |
+| **save_meta**                            | **save_metadata_to_images**          | name change only      |
+| **meta_scheme**                          | **metadata_scheme**                  | name change only      |
+| **save_extension**                       | **output_format**                    | name change only      |
+| save_name                                |                                      | remove        |
+| read_wildcards_in_order                  | read_wildcards_in_order              |                     |
+| require_base64                           | require_base64                       | will be remove         |
+| async_process                            | async_process                        |                     |
+| webhook_url                              | webhook_url                          |                     |
+
+simple is:
+
+- All `AdvancedParams` move to upper level
+- Modify some params name
+    - `input_image` -> `inpaint_input_image`
+    - `inpaint_mask` -> `inpaint_mask_image_upload`
+    - `input_image` -> `uov_input_image`
+    - `image_prompts` -> `controlnet_image`
+    - `upscale_value` -> `upscale_value`
+    - `save_meta` -> `upscale_multiple`
+    - `meta_scheme` -> `save_metadata_to_images`
+    - `save_extension` -> `output_format`
+- Remove some params
+    - `save_name`
+- Add some params
+    - `input_image_checkbox`
+    - `current_tab`
+    - `generate_image_grid`
+    - `preset`
+    - `stream_output`
+- Merge some params
+    - `outpaint_distance_left,right,top,bottom` 四个参数合并为 `outpaint_distance`
+
+## Example for three types of return
+
+### async task
+
+specify `async_process` as `True`
+
+```python
+import requests
+import json
+
+endpoint = "http://127.0.0.1:7866/v1/engine/generate/"
+
+params = {
+    "prompt": "",
+    "negative_prompt": "",
+    "performance_selection": "Lightning",
+    "async_process": True,
+    "webhook_url": ""
+}
+
+res = requests.post(
+    url=endpoint,
+    data=json.dumps(params),
+    timeout=60
+)
+
+print(res.json())
+```
+
+output will be like this:
+
+```python
+{'id': -1, 'task_id': '85c10c81e9e2482d90a64c3704137d3a', 'req_params': {}, 'in_queue_mills': -1, 'start_mills': -1, 'finish_mills': -1, 'task_status': 'pending', 'progress': -1, 'preview': '', 'webhook_url': '', 'result': []}
+```
+
+use `task_id` request `http://127.0.0.1:7866/tasks/{task_id}` to get task info, if this task is currently running, return should be include `preview`
+
+example for return
+
+```python
+# pending
+{
+    "id": -1,
+    "in_queue_mills": 1720085748199,
+    "finish_mills": null,
+    "progress": null,
+    "result": null,
+    "req_params": {
+        # full request params
+        ...
+    },
+    "task_id": "85c10c81e9e2482d90a64c3704137d3a",
+    "start_mills": null,
+    "task_status": null,
+    "webhook_url": ""
+}
+
+# running
+{
+    "id": -1,
+    "task_id": "85c10c81e9e2482d90a64c3704137d3a",
+    "req_params": {
+        ...
+    },
+    "in_queue_mills": 1720086131653,
+    "start_mills": 1720086131865,
+    "finish_mills": -1,
+    "task_status": "running",
+    "progress": 18,
+    "preview": "a long text",
+    "webhook_url": "",
+    "result": []
+}
+
+# finished
+{
+    "id": 71,
+    "in_queue_mills": 1720085748199,
+    "finish_mills": 1720085770046,
+    "progress": 100,
+    "result": [
+        "http://127.0.0.1:7866/outputs/2024-07-04/2024-07-04_17-36-09_5201.png"
+    ],
+    "req_params": {
+        ...
+    },
+    "task_id": "85c10c81e9e2482d90a64c3704137d3a",
+    "start_mills": 1720085748425,
+    "task_status": "finished",
+    "webhook_url": ""
+}
+```
+
+### streaming output
+
+this is like LLM streaming output, you will recieve from server until finish, refer to the above example:
+
+```python
+import requests
+import json
+
+endpoint = "http://127.0.0.1:7866/v1/engine/generate/"
+
+params = {
+    "prompt": "",
+    "negative_prompt": "",
+    "performance_selection": "Lightning",
+    "stream_output": True,
+    "webhook_url": ""
+}
+
+res = requests.post(
+    url=endpoint,
+    data=json.dumps(params),
+    stream=True,
+    timeout=60
+)
+
+for line in res.iter_lines():
+    if line:
+        print(line.decode('utf-8'))
+```
+
+you will get response like this:
+
+```python
+data: {"progress": 2, "preview": null, "message": "Loading models ...", "images": []}
+data:
+data: {"progress": 13, "preview": null, "message": "Preparing task 1/1 ...", "images": []}
+data:
+data: {"progress": 13, "preview": "data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAASAAAA...", 'message': 'Sampling step 1/4, image 1/1 ...', 'images': []}
+data:
+data: {"progress": 34, "preview": "data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAASAAAA...", 'message': 'Sampling step 2/4, image 1/1 ...', 'images': []}
+data:
+data: {"progress": 56, "preview": "data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAASAAAA...", 'message': 'Sampling step 3/4, image 1/1 ...', 'images': []}
+data:
+data: {"progress": 78, "preview": "data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAASAAAA...", 'message': 'Sampling step 4/4, image 1/1 ...', 'images': []}
+data:
+data: {"progress": 100, "preview": null, "message": "Saving image 1/1 to system ...", "images": []}
+data:
+data: {"progress": 100, "preview": null, "message": "Finished", "images": ["http://10.0.0.245:7866/outputs/2024-07-05/2024-07-05_09-31-10_1752.png"]}
+data:
+```
+
+just modify our code:
+
+```python
+import requests
+import json
+
+endpoint = "http://127.0.0.1:7866/v1/engine/generate/"
+
+params = {
+    "prompt": "",
+    "negative_prompt": "",
+    "performance_selection": "Lightning",
+    "stream_output": True,
+    "webhook_url": ""
+}
+
+res = requests.post(
+    url=endpoint,
+    data=json.dumps(params),
+    stream=True,
+    timeout=60
+)
+
+for line in res.iter_lines(chunk_size=8192):
+    line = line.decode('utf-8').split('\n')[0]
+
+    try:
+        json_data = json.loads(line[6:])
+        if json_data["preview"] is not None:
+            json_data["preview"] = "data:image/png;base64,iVBORw0KGgoAAAANSU..."
+    except json.decoder.JSONDecodeError:
+        continue
+    print(json_data)
+```
+
+you will get this:
+
+```python
+{'progress': 13, 'preview': None, 'message': 'Preparing task 1/1 ...', 'images': []}
+{'progress': 13, 'preview': 'data:image/png;base64,iVBORw0KGgoAAAANSU...', 'message': 'Sampling step 1/4, image 1/1 ...', 'images': []}
+{'progress': 34, 'preview': 'data:image/png;base64,iVBORw0KGgoAAAANSU...', 'message': 'Sampling step 2/4, image 1/1 ...', 'images': []}
+{'progress': 56, 'preview': 'data:image/png;base64,iVBORw0KGgoAAAANSU...', 'message': 'Sampling step 3/4, image 1/1 ...', 'images': []}
+{'progress': 78, 'preview': 'data:image/png;base64,iVBORw0KGgoAAAANSU...', 'message': 'Sampling step 4/4, image 1/1 ...', 'images': []}
+{'progress': 100, 'preview': None, 'message': 'Saving image 1/1 to system ...', 'images': []}
+{'progress': 100, 'preview': None, 'message': 'Finished', 'images': ['http://10.0.0.245:7866/outputs/2024-07-05/2024-07-05_10-02-22_2536.png']}
+```
+
+it is better for frontend i think (but i am not good at this). with AI, i generate a [example.html](./docs/example.html), click `Generate` button, you will get a page with preview and progress.
+
+### binary output
+
+this is simple, return is a image, pass `async_process` and `stream_output` both `false`, at this time, `image_number` force to `1`
+
+```python
+import requests
+import json
+from PIL import Image
+from io import BytesIO
+import matplotlib.pyplot as plt
+
+endpoint = "http://127.0.0.1:7866/v1/engine/generate/"
+
+params = {
+    "prompt": "",
+    "negative_prompt": "",
+    "performance_selection": "Lightning",
+    "async_process": False,
+    "stream_output": False,
+    "webhook_url": ""
+}
+
+res = requests.post(
+    url=endpoint,
+    data=json.dumps(params),
+    timeout=60
+)
+
+image_stream = BytesIO(res.content)
+image = Image.open(image_stream)
+
+plt.imshow(image)
+plt.show()
+```
+
+# task query
+
+Unlike [Fooocus-API](https://github.com/mrhan1993/Fooocus-API), the history saving will be automatic without a retention switch. The database is used with SQLite3 and stored in `outputs/db.sqlite3`. Taking lessons from the previous version, the table structure has been greatly simplified, and request parameters are stored as JSON in the `req_params` field. To reduce read and write operations, database operations are only performed when tasks enter and complete the queue. It is only used for generating records, and task status tracking is completed in memory.
+
+In addition, this version will retain input images, uploaded images will calculate hash values and be saved in the `inputs` directory, and the image parameters in the database's `req_params` will be replaced with `url` information for saving, which means more complete historical record preservation, whether it is text-to-image or image-to-image or other types of images.
+
+## /tasks
+
+This is a compound interface, but its return format is fixed. The interface will always return JSON data in the following format, regardless of how the parameters are specified.
+
+```python
+{
+    "history": [],
+    "current": [],  # Although it is a list, there will be no more than one element in it.
+    "pending": []
+}
+```
+
+All elements have a format that matches the scheme in the database, except for current which has an additional preview, as shown in the following figure:
+
+![](./assets/tasks.png)
+
+more usage, see below:
+
+> The return format of this interface is always fixed, regardless of how the parameters are adjusted.
+
+```shell
+curl http://localhost:7866/tasks?query=current
+# only return current task, other value for query include 'all', 'pending', 'history'
+
+curl http://localhost:7866/tasks?query=history&page=3&page_size=5
+# history and pending supports pagination and page size.
+
+curl http://localhost:7866/tasks?query=history&start_at=2024-07-03T12:22:30
+# You can specify a time range for the query, which will return all records within that time period. The time format is ISO8601, and if you do not specify end_at, it will be set to the current time.
+
+curl http://localhost:7866/tasks?query=history&start_at=2024-07-03T12:22:30&action=delete
+# Delete tasks within a specified time range, including database records and generated files. This is the only supported deletion method at present (input files will not be deleted).
+
+curl http://localhost:7866/tasks/38ba92b188a64233a7336218cd902865
+# This will return the information of the task, but it is just a dictionary. It is equivalent to taking the task with the specified task_id from the list above. If it happens to be the current task, it will also include preview. (Although it may look similar, this is actually another interface.)
+```

docs/migrate_zh.md

@@ -0,0 +1,377 @@
+# 参数对照表
+
+`AdvancedParams` 用 `adp` 代替，名称变动的原则是和 Fooocus 进行统一：
+
+| Fooocus-API                              | FooocusAPI                           | 备注                  |
+|------------------------------------------|--------------------------------------|---------------------|
+| prompt                                   | prompt                               |                     |
+| negative_prompt                          | negative_prompt                      |                     |
+| style_selections                         | style_selections                     |                     |
+| performance_selection                    | performance_selection                |                     |
+| aspect_ratios_selection                  | aspect_ratios_selection              |                     |
+| image_number                             | image_number                         |                     |
+| image_seed                               | image_seed                           |                     |
+| sharpness                                | sharpness                            |                     |
+| guidance_scale                           | guidance_scale                       |                     |
+| base_model_name                          | base_model_name                      |                     |
+| refiner_model_name                       | refiner_model_name                   |                     |
+| refiner_switch                           | refiner_switch                       |                     |
+| loras                                    | loras                                | 传入格式相同，都是 Lora 对象列表 |
+|                                          | input_image_checkbox                 | 可以忽略，它总是为 True      |
+|                                          | current_tab                          | 可以忽略，根据参数会自动判断      |
+| uov_method                               | uov_method                           |                     |
+| **input_image**                          | **uov_input_image**                  | 使用 Fooocus 的变量名称    |
+| outpaint_selections                      | outpaint_selections                  |                     |
+| **input_image**                          | **inpaint_input_image**              | 使用 Fooocus 的变量名称    |
+| inpaint_additional_prompt                | inpaint_additional_prompt            |                     |
+| **input_mask**                           | **inpaint_mask_image_upload**        | 使用 Fooocus 的变量名称    |
+| adp.disable_preview                      | disable_preview                      |                     |
+| adp.disable_intermediate_results         | disable_intermediate_results         |                     |
+| adp.disable_seed_increment               | disable_seed_increment               |                     |
+| adp.black_out_nsfw                       | black_out_nsfw                       |                     |
+| adp.adm_scaler_positive                  | adm_scaler_positive                  |                     |
+| adp.adm_scaler_negative                  | adm_scaler_negative                  |                     |
+| adp.adm_scaler_end                       | adm_scaler_end                       |                     |
+| adp.adaptive_cfg                         | adaptive_cfg                         |                     |
+| adp.clip_skip                            | clip_skip                            |                     |
+| adp.sampler_name                         | sampler_name                         |                     |
+| adp.scheduler_name                       | scheduler_name                       |                     |
+| adp.vae_name                             | vae_name                             |                     |
+| adp.overwrite_step                       | overwrite_step                       |                     |
+| adp.overwrite_switch                     | overwrite_switch                     |                     |
+| adp.overwrite_width                      | overwrite_width                      |                     |
+| adp.overwrite_height                     | overwrite_height                     |                     |
+| adp.overwrite_vary_strength              | overwrite_vary_strength              |                     |
+| adp.overwrite_upscale_strength           | overwrite_upscale_strength           |                     |
+| adp.mixing_image_prompt_and_vary_upscale | mixing_image_prompt_and_vary_upscale |                     |
+| adp.mixing_image_prompt_and_inpaint      | mixing_image_prompt_and_inpaint      |                     |
+| adp.debugging_cn_preprocessor            | debugging_cn_preprocessor            |                     |
+| adp.skipping_cn_preprocessor             | skipping_cn_preprocessor             |                     |
+| adp.canny_low_threshold                  | canny_low_threshold                  |                     |
+| adp.canny_high_threshold                 | canny_high_threshold                 |                     |
+| adp.refiner_swap_method                  | refiner_swap_method                  |                     |
+| adp.controlnet_softness                  | controlnet_softness                  |                     |
+| adp.freeu_enabled                        | freeu_enabled                        |                     |
+| adp.freeu_b1                             | freeu_b1                             |                     |
+| adp.freeu_b2                             | freeu_b2                             |                     |
+| adp.freeu_s1                             | freeu_s1                             |                     |
+| adp.freeu_s2                             | freeu_s2                             |                     |
+| adp.debugging_inpaint_preprocessor       | debugging_inpaint_preprocessor       |                     |
+| adp.inpaint_disable_initial_latent       | inpaint_disable_initial_latent       |                     |
+| adp.inpaint_engine                       | inpaint_engine                       |                     |
+| adp.inpaint_strength                     | inpaint_strength                     |                     |
+| adp.inpaint_respective_field             | inpaint_respective_field             |                     |
+| adp.inpaint_mask_upload_checkbox         | inpaint_mask_upload_checkbox         |                     |
+| adp.invert_mask_checkbox                 | invert_mask_checkbox                 |                     |
+| adp.inpaint_erode_or_dilate              | inpaint_erode_or_dilate              |                     |
+| **image_prompts**                        | **controlnet_image**                 | 只是属性名称变更            |
+|                                          | generate_image_grid                  | 新增，这是个测试选项，建议默认     |
+| outpaint_distance_left                   | outpaint_distance                    | 这四个属性合并为了一个属性       |
+| outpaint_distance_right                  |                                      | 可以通过一个列表传递这四个值      |
+| outpaint_distance_top                    |                                      | 例如：[100, 50, 0, 0]  |
+| outpaint_distance_bottom                 |                                      | 方向是：左, 上, 右, 下      |
+| **upscale_value**                        | **upscale_multiple**                 | 属性名变更               |
+|                                          | preset                               | 新增，可以通过该属性指定使用的预设   |
+|                                          | stream_output                        | 新增流式输出，类似 LLM 的流式输出 |
+| **save_meta**                            | **save_metadata_to_images**          |                     |
+| **meta_scheme**                          | **metadata_scheme**                  |                     |
+| **save_extension**                       | **output_format**                    |                     |
+| save_name                                |                                      | 移除，不支持自定义文件名        |
+| read_wildcards_in_order                  | read_wildcards_in_order              |                     |
+| require_base64                           | require_base64                       | 该参数后续可能会被移除         |
+| async_process                            | async_process                        |                     |
+| webhook_url                              | webhook_url                          |                     |
+
+简单说来就是
+
+- 将所有 `AdvancedParams` 平移到上一级
+- 修改部分参数名
+    - `input_image` -> `inpaint_input_image`
+    - `inpaint_mask` -> `inpaint_mask_image_upload`
+    - `input_image` -> `uov_input_image`
+    - `image_prompts` -> `controlnet_image`
+    - `upscale_value` -> `upscale_value`
+    - `save_meta` -> `upscale_multiple`
+    - `meta_scheme` -> `save_metadata_to_images`
+    - `save_extension` -> `output_format`
+- 移除部分参数名
+    - `save_name`
+- 增加部分参数
+    - `input_image_checkbox`
+    - `current_tab`
+    - `generate_image_grid`
+    - `preset`
+    - `stream_output`
+- 合并部分参数
+    - `outpaint_distance_left,right,top,bottom` 四个参数合并为 `outpaint_distance`
+
+## 三种返回示例
+
+### 异步任务
+
+在参数中指定 `async_process` 为 `True`
+
+```python
+import requests
+import json
+
+endpoint = "http://127.0.0.1:7866/v1/engine/generate/"
+
+params = {
+    "prompt": "",
+    "negative_prompt": "",
+    "performance_selection": "Lightning",
+    "async_process": True,
+    "webhook_url": ""
+}
+
+res = requests.post(
+    url=endpoint,
+    data=json.dumps(params),
+    timeout=60
+)
+
+print(res.json())
+```
+
+输出如下：
+
+```python
+{'id': -1, 'task_id': '85c10c81e9e2482d90a64c3704137d3a', 'req_params': {}, 'in_queue_mills': -1, 'start_mills': -1, 'finish_mills': -1, 'task_status': 'pending', 'progress': -1, 'preview': '', 'webhook_url': '', 'result': []}
+```
+
+你可以通过 `task_id` 访问 `http://127.0.0.1:7866/tasks/{task_id}` 获取任务信息，如果该任务正在执行，返回信息中会包含 `preview`
+
+返回数据示例：
+
+```python
+# 未开始
+{
+    "id": -1,
+    "in_queue_mills": 1720085748199,
+    "finish_mills": null,
+    "progress": null,
+    "result": null,
+    "req_params": {
+        # 完整的请求参数
+        ...
+    },
+    "task_id": "85c10c81e9e2482d90a64c3704137d3a",
+    "start_mills": null,
+    "task_status": null,
+    "webhook_url": ""
+}
+
+# 执行中
+{
+    "id": -1,
+    "task_id": "85c10c81e9e2482d90a64c3704137d3a",
+    "req_params": {
+        ...
+    },
+    "in_queue_mills": 1720086131653,
+    "start_mills": 1720086131865,
+    "finish_mills": -1,
+    "task_status": "running",
+    "progress": 18,
+    "preview": "a long text",
+    "webhook_url": "",
+    "result": []
+}
+
+# 已完成
+{
+    "id": 71,
+    "in_queue_mills": 1720085748199,
+    "finish_mills": 1720085770046,
+    "progress": 100,
+    "result": [
+        "http://127.0.0.1:7866/outputs/2024-07-04/2024-07-04_17-36-09_5201.png"
+    ],
+    "req_params": {
+        ...
+    },
+    "task_id": "85c10c81e9e2482d90a64c3704137d3a",
+    "start_mills": 1720085748425,
+    "task_status": "finished",
+    "webhook_url": ""
+}
+```
+
+### 流式输出
+
+这是一个类似 LLM 流式输出的方式，你会持续收到来自服务器的信息，直到结束，参照上面的示例：
+
+```python
+import requests
+import json
+
+endpoint = "http://127.0.0.1:7866/v1/engine/generate/"
+
+params = {
+    "prompt": "",
+    "negative_prompt": "",
+    "performance_selection": "Lightning",
+    "stream_output": True,
+    "webhook_url": ""
+}
+
+res = requests.post(
+    url=endpoint,
+    data=json.dumps(params),
+    stream=True,
+    timeout=60
+)
+
+for line in res.iter_lines():
+    if line:
+        print(line.decode('utf-8'))
+```
+
+你会获得类似下面的输出：
+
+```python
+data: {"progress": 2, "preview": null, "message": "Loading models ...", "images": []}
+data:
+data: {"progress": 13, "preview": null, "message": "Preparing task 1/1 ...", "images": []}
+data:
+data: {"progress": 13, "preview": "data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAASAAAA...", 'message': 'Sampling step 1/4, image 1/1 ...', 'images': []}
+data:
+data: {"progress": 34, "preview": "data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAASAAAA...", 'message': 'Sampling step 2/4, image 1/1 ...', 'images': []}
+data:
+data: {"progress": 56, "preview": "data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAASAAAA...", 'message': 'Sampling step 3/4, image 1/1 ...', 'images': []}
+data:
+data: {"progress": 78, "preview": "data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAASAAAA...", 'message': 'Sampling step 4/4, image 1/1 ...', 'images': []}
+data:
+data: {"progress": 100, "preview": null, "message": "Saving image 1/1 to system ...", "images": []}
+data:
+data: {"progress": 100, "preview": null, "message": "Finished", "images": ["http://10.0.0.245:7866/outputs/2024-07-05/2024-07-05_09-31-10_1752.png"]}
+data:
+```
+
+我们在稍微修改下：
+
+```python
+import requests
+import json
+
+endpoint = "http://127.0.0.1:7866/v1/engine/generate/"
+
+params = {
+    "prompt": "",
+    "negative_prompt": "",
+    "performance_selection": "Lightning",
+    "stream_output": True,
+    "webhook_url": ""
+}
+
+res = requests.post(
+    url=endpoint,
+    data=json.dumps(params),
+    stream=True,
+    timeout=60
+)
+
+for line in res.iter_lines(chunk_size=8192):
+    line = line.decode('utf-8').split('\n')[0]
+
+    try:
+        json_data = json.loads(line[6:])
+        if json_data["preview"] is not None:
+            json_data["preview"] = "data:image/png;base64,iVBORw0KGgoAAAANSU..."
+    except json.decoder.JSONDecodeError:
+        continue
+    print(json_data)
+```
+
+然后你就得到了一系列类似这样的输出：
+
+```python
+{'progress': 13, 'preview': None, 'message': 'Preparing task 1/1 ...', 'images': []}
+{'progress': 13, 'preview': 'data:image/png;base64,iVBORw0KGgoAAAANSU...', 'message': 'Sampling step 1/4, image 1/1 ...', 'images': []}
+{'progress': 34, 'preview': 'data:image/png;base64,iVBORw0KGgoAAAANSU...', 'message': 'Sampling step 2/4, image 1/1 ...', 'images': []}
+{'progress': 56, 'preview': 'data:image/png;base64,iVBORw0KGgoAAAANSU...', 'message': 'Sampling step 3/4, image 1/1 ...', 'images': []}
+{'progress': 78, 'preview': 'data:image/png;base64,iVBORw0KGgoAAAANSU...', 'message': 'Sampling step 4/4, image 1/1 ...', 'images': []}
+{'progress': 100, 'preview': None, 'message': 'Saving image 1/1 to system ...', 'images': []}
+{'progress': 100, 'preview': None, 'message': 'Finished', 'images': ['http://10.0.0.245:7866/outputs/2024-07-05/2024-07-05_10-02-22_2536.png']}
+```
+
+这还挺适合前端套壳用的（可惜我完全搞不懂前端，要不高低套一个），比如��用 AI 生成了一个 [example.html](./docs/example.html) ，服务启动后点击 `Generate` 按钮，你就会得到一个有预览、有进度的生成过程。
+
+### 二进制输出
+
+这个就简单了，它就是返回一张图片，不过需要在请求时将 `async_process` 和 `stream_output` 同时指定为 `false`，此时 `image_number` 强制为 `1`
+
+```python
+import requests
+import json
+from PIL import Image
+from io import BytesIO
+import matplotlib.pyplot as plt
+
+endpoint = "http://127.0.0.1:7866/v1/engine/generate/"
+
+params = {
+    "prompt": "",
+    "negative_prompt": "",
+    "performance_selection": "Lightning",
+    "async_process": False,
+    "stream_output": False,
+    "webhook_url": ""
+}
+
+res = requests.post(
+    url=endpoint,
+    data=json.dumps(params),
+    timeout=60
+)
+
+image_stream = BytesIO(res.content)
+image = Image.open(image_stream)
+
+plt.imshow(image)
+plt.show()
+```
+
+# 任务查询
+
+和 [Fooocus-API](https://github.com/mrhan1993/Fooocus-API) 不同的是历史记录的保存将是自动进行的，没有保留开关。数据库使用 `SQLite3` 并存放在 `outputs/db.sqlite3` 中。同时吸取了上次的教训，极大简化了表结构，将请求参数作为 JSON 存放在 `req_params` 字段。为了降低读写，仅在任务进入队列时和完成后进行数据库操作。其仅作为生成记录使用，任务状态的追踪会在内存中完成。
+
+此外，该版本会保留输入图像，上传的图像会计算哈希值并保存在 `inputs` 目录，数据库中的 `req_params` 会将图片参数替换为 `url` 信息进行保存，这意味着更完整的历史记录保存，无论是文生图还是图生图又或者是其他
+
+## /tasks
+
+这是个复合接口，但其返回格式是固定的，该接口总是会返回下面格式的 JSON 数据，无论参数如何指定
+
+```python
+{
+    "history": [],
+    "current": [],  # 尽管是个列表，但其中不会超过一个元素。
+    "pending": []
+}
+```
+
+所有的元素其格式都是和数据库中的 scheme 匹配的，除了 `current` 会多一个 `preview` ，比如下图：
+
+![](./assets/tasks.png)
+
+该接口还支持更加精细的用法，参考下面的示例：
+
+> 该接口返回格式总是固定的，不管参数如何调整
+
+```shell
+curl http://localhost:7866/tasks?query=current
+# 仅返回当前任务，query 参数还可以指定的值为 'all', 'pending', 'history'
+
+curl http://localhost:7866/tasks?query=history&page=3&page_size=5
+# history 和 pending 支持分页和页面大小
+
+curl http://localhost:7866/tasks?query=history&start_at=2024-07-03T12:22:30
+# 你可以指定一个时间范围进行查询，这会返回该时间段的所有记录。时间格式是 ISO8601，如果你不指定 end_at 则截止当前时间
+
+curl http://localhost:7866/tasks?query=history&start_at=2024-07-03T12:22:30&action=delete
+# 删除指定时间范围的任务，数据库记录和生成文件。目前仅支持这一种删除方法(不会删除 input 文件)。
+
+curl http://localhost:7866/tasks/38ba92b188a64233a7336218cd902865
+# 这会返回该任务的信息，但它只是一个字典。相当于从上面列表中取出指定 task_id 的任务，如果它刚好是当前任务，那它也会包含 preview
+```

environment.yaml

@@ -0,0 +1,7 @@
+name: fooocus-api
+channels:
+  - defaults
+dependencies:
+  - python=3.10
+  - pip=23.0
+  - packaging

examples/Note.txt

@@ -0,0 +1,3 @@
+All image in imgs folder are generated by AI, so there is no copyright issue.
+
+所有 imgs 目录中的图片都是 AI 生成的，不存在版权问题

examples/examples.ipynb

@@ -0,0 +1,521 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# text to image"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-04-04T10:19:57.369099Z",
+     "start_time": "2024-04-04T10:19:57.328298Z"
+    }
+   },
+   "outputs": [],
+   "source": [
+    "import requests\n",
+    "import json\n",
+    "\n",
+    "# Vincent diagram example\n",
+    "host = \"http://127.0.0.1:8888\"\n",
+    "\n",
+    "def text2img(params: dict) -> dict:\n",
+    "    \"\"\"\n",
+    "    Vincent picture\n",
+    "    \"\"\"\n",
+    "    response = requests.post(\n",
+    "        url=f\"{host}/v1/generation/text-to-image\",\n",
+    "        data=json.dumps(params),\n",
+    "        headers={\"Content-Type\": \"application/json\"})\n",
+    "    return response.json()\n",
+    "\n",
+    "result =text2img({\n",
+    "    \"performance_selection\": \"Lightning\",\n",
+    "    \"async_process\": True\n",
+    "})\n",
+    "print(result)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# upscale or vary"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import requests\n",
+    "import json\n",
+    "\n",
+    "\n",
+    "# upscale or vary v1 Interface example\n",
+    "host = \"http://127.0.0.1:8888\"\n",
+    "image = open(\"./imgs/bear.jpg\", \"rb\").read()\n",
+    "\n",
+    "def upscale_vary(image, params: dict) -> dict:\n",
+    "    \"\"\"\n",
+    "    Upscale or Vary\n",
+    "    \"\"\"\n",
+    "    response = requests.post(\n",
+    "        url=f\"{host}/v1/generation/image-upscale-vary\",\n",
+    "        data=params,\n",
+    "        files={\"input_image\": image})\n",
+    "    return response.json()\n",
+    "\n",
+    "result =upscale_vary(\n",
+    "    image=image,\n",
+    "    params={\n",
+    "        \"uov_method\": \"Vary\",\n",
+    "        \"async_process\": True\n",
+    "    })\n",
+    "print(json.dumps(result, indent=4, ensure_ascii=False))"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import requests\n",
+    "import json\n",
+    "import base64\n",
+    "\n",
+    "\n",
+    "# upscale or vary v2 Interface example\n",
+    "host = \"http://127.0.0.1:8888\"\n",
+    "image = open(\"./imgs/bear.jpg\", \"rb\").read()\n",
+    "\n",
+    "def upscale_vary(params: dict) -> dict:\n",
+    "    \"\"\"\n",
+    "    Upscale or Vary\n",
+    "    \"\"\"\n",
+    "    response = requests.post(\n",
+    "        url=f\"{host}/v2/generation/image-upscale-vary\",\n",
+    "        data=json.dumps(params),\n",
+    "        headers={\"Content-Type\": \"application/json\"},\n",
+    "        timeout=300)\n",
+    "    return response.json()\n",
+    "\n",
+    "result =upscale_vary(\n",
+    "    params={\n",
+    "        \"input_image\": base64.b64encode(image).decode('utf-8'),\n",
+    "        \"uov_method\": \"Upscale (2x)\",\n",
+    "        \"async_process\": True\n",
+    "    })\n",
+    "print(json.dumps(result, indent=4, ensure_ascii=False))"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# inpaint or outpaint"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import requests\n",
+    "import json\n",
+    "\n",
+    "# Partial redraw v1 interface example\n",
+    "host = \"http://127.0.0.1:8888\"\n",
+    "image = open(\"./imgs/bear.jpg\", \"rb\").read()\n",
+    "\n",
+    "def inpaint_outpaint(params: dict, input_image: bytes, input_mask: bytes = None) -> dict:\n",
+    "    \"\"\"\n",
+    "    Partial redraw v1 interface example\n",
+    "    \"\"\"\n",
+    "    response = requests.post(\n",
+    "        url=f\"{host}/v1/generation/image-inpaint-outpaint\",\n",
+    "        data=params,\n",
+    "        files={\"input_image\": input_image,\n",
+    "        \"input_mask\": input_mask})\n",
+    "    return response.json()\n",
+    "\n",
+    "\n",
+    "# Image extension example\n",
+    "result = inpaint_outpaint(\n",
+    "    params={\n",
+    "        \"outpaint_selections\": \"Left,Right\",\n",
+    "        \"async_process\": True},\n",
+    "    input_image=image,\n",
+    "    input_mask=None)\n",
+    "print(json.dumps(result, indent=4, ensure_ascii=False))"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "#Partial redraw example\n",
+    "source = open(\"./imgs/inpaint_source.jpg\", \"rb\").read()\n",
+    "mask = open(\"./imgs/inpaint_mask.png\", \"rb\").read()\n",
+    "result = inpaint_outpaint(\n",
+    "    params={\n",
+    "        \"prompt\": \"a cat\",\n",
+    "        \"async_process\": True\n",
+    "    },\n",
+    "    input_image=source,\n",
+    "    input_mask=mask)\n",
+    "print(json.dumps(result, indent=4, ensure_ascii=False))"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import requests\n",
+    "import json\n",
+    "import base64\n",
+    "\n",
+    "\n",
+    "# Partial redraw v2 interface example\n",
+    "host = \"http://127.0.0.1:8888\"\n",
+    "image = open(\"./imgs/bear.jpg\", \"rb\").read()\n",
+    "\n",
+    "def inpaint_outpaint(params: dict) -> dict:\n",
+    "    \"\"\"\n",
+    "    Partial redraw v2 interface example\n",
+    "    \"\"\"\n",
+    "    response = requests.post(\n",
+    "        url=f\"{host}/v2/generation/image-inpaint-outpaint\",\n",
+    "        data=json.dumps(params),\n",
+    "        headers={\"Content-Type\": \"application/json\"})\n",
+    "    return response.json()\n",
+    "\n",
+    "# Image extension example\n",
+    "result = inpaint_outpaint(\n",
+    "    params={\n",
+    "        \"input_image\": base64.b64encode(image).decode('utf-8'),\n",
+    "        \"input_mask\": None,\n",
+    "        \"outpaint_selections\": [\"Left\", \"Right\"],\n",
+    "        \"async_process\": True\n",
+    "    })\n",
+    "print(json.dumps(result, indent=4, ensure_ascii=False))"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Partial redraw example\n",
+    "source = open(\"./imgs/inpaint_source.jpg\", \"rb\").read()\n",
+    "mask = open(\"./imgs/inpaint_mask.png\", \"rb\").read()\n",
+    "result = inpaint_outpaint(\n",
+    "    params={\n",
+    "        \"prompt\": \"a cat\",\n",
+    "        \"input_image\": base64.b64encode(source).decode('utf-8'),\n",
+    "        \"input_mask\": base64.b64encode(mask).decode('utf-8'),\n",
+    "        \"async_process\": True\n",
+    "    })\n",
+    "print(json.dumps(result, indent=4, ensure_ascii=False))"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# image prompts"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import requests\n",
+    "import json\n",
+    "\n",
+    "\n",
+    "# image_prompt v1 Interface example\n",
+    "host = \"http://127.0.0.1:8888\"\n",
+    "image = open(\"./imgs/bear.jpg\", \"rb\").read()\n",
+    "source = open(\"./imgs/inpaint_source.jpg\", \"rb\").read()\n",
+    "mask = open(\"./imgs/inpaint_mask.png\", \"rb\").read()\n",
+    "\n",
+    "def image_prompt(\n",
+    "        params: dict,\n",
+    "        input_image: bytes=None,\n",
+    "        input_mask: bytes=None,\n",
+    "        cn_img1: bytes=None,\n",
+    "        cn_img2: bytes=None,\n",
+    "        cn_img3: bytes=None,\n",
+    "        cn_img4: bytes=None,) -> dict:\n",
+    "    \"\"\"\n",
+    "    image prompt\n",
+    "    \"\"\"\n",
+    "    response = requests.post(\n",
+    "        url=f\"{host}/v1/generation/image-prompt\",\n",
+    "        data=params,\n",
+    "        files={\n",
+    "            \"input_image\": input_image,\n",
+    "            \"input_mask\": input_mask,\n",
+    "            \"cn_img1\": cn_img1,\n",
+    "            \"cn_img2\": cn_img2,\n",
+    "            \"cn_img3\": cn_img3,\n",
+    "            \"cn_img4\": cn_img4,\n",
+    "        })\n",
+    "    return response.json()\n",
+    "\n",
+    "# image extension\n",
+    "params = {\n",
+    "    \"outpaint_selections\": [\"Left\", \"Right\"],\n",
+    "    \"image_prompts\": [] # Required parameters, can be an empty list\n",
+    "}\n",
+    "result = image_prompt(params=params, input_image=image)\n",
+    "print(json.dumps(result, indent=4, ensure_ascii=False))"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# partial redraw\n",
+    "\n",
+    "params = {\n",
+    "    \"prompt\": \"1girl sitting on the chair\",\n",
+    "    \"image_prompts\": [], # Required parameters, can be an empty list\n",
+    "    \"async_process\": True\n",
+    "}\n",
+    "result = image_prompt(params=params, input_image=source, input_mask=mask)\n",
+    "print(json.dumps(result, indent=4, ensure_ascii=False))"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# image prompt\n",
+    "\n",
+    "params = {\n",
+    "    \"prompt\": \"1girl sitting on the chair\",\n",
+    "    \"image_prompts\": [\n",
+    "        {\n",
+    "            \"cn_stop\": 0.6,\n",
+    "            \"cn_weight\": 0.6,\n",
+    "            \"cn_type\": \"ImagePrompt\"\n",
+    "        },{\n",
+    "            \"cn_stop\": 0.6,\n",
+    "            \"cn_weight\": 0.6,\n",
+    "            \"cn_type\": \"ImagePrompt\"\n",
+    "        }]\n",
+    "    }\n",
+    "result = image_prompt(params=params, cn_img1=image, cn_img2=source)\n",
+    "print(json.dumps(result, indent=4, ensure_ascii=False))"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import requests\n",
+    "import json\n",
+    "import base64\n",
+    "\n",
+    "# image_prompt v2 Interface example\n",
+    "host = \"http://127.0.0.1:8888\"\n",
+    "image = open(\"./imgs/bear.jpg\", \"rb\").read()\n",
+    "source = open(\"./imgs/inpaint_source.jpg\", \"rb\").read()\n",
+    "mask = open(\"./imgs/inpaint_mask.png\", \"rb\").read()\n",
+    "\n",
+    "def image_prompt(params: dict) -> dict:\n",
+    "    \"\"\"\n",
+    "    image prompt\n",
+    "    \"\"\"\n",
+    "    response = requests.post(\n",
+    "        url=f\"{host}/v2/generation/image-prompt\",\n",
+    "        data=json.dumps(params),\n",
+    "        headers={\"Content-Type\": \"application/json\"})\n",
+    "    return response.json()\n",
+    "\n",
+    "# image extension\n",
+    "params = {\n",
+    "    \"input_image\": base64.b64encode(image).decode('utf-8'),\n",
+    "    \"outpaint_selections\": [\"Left\", \"Right\"],\n",
+    "    \"image_prompts\": [] # Required parameters, can be an empty list\n",
+    "}\n",
+    "result = image_prompt(params)\n",
+    "print(json.dumps(result, indent=4, ensure_ascii=False))"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# partial redraw\n",
+    "\n",
+    "params = {\n",
+    "    \"prompt\": \"1girl sitting on the chair\",\n",
+    "    \"input_image\": base64.b64encode(source).decode('utf-8'),\n",
+    "    \"input_mask\": base64.b64encode(mask).decode('utf-8'),\n",
+    "    \"image_prompts\": [], # Required parameters, can be an empty list\n",
+    "    \"async_process\": True\n",
+    "}\n",
+    "result = image_prompt(params)\n",
+    "print(json.dumps(result, indent=4, ensure_ascii=False))"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# image prompt\n",
+    "\n",
+    "params = {\n",
+    "    \"prompt\": \"1girl sitting on the chair\",\n",
+    "    \"image_prompts\": [\n",
+    "        {\n",
+    "            \"cn_img\": base64.b64encode(source).decode('utf-8'),\n",
+    "            \"cn_stop\": 0.6,\n",
+    "            \"cn_weight\": 0.6,\n",
+    "            \"cn_type\": \"ImagePrompt\"\n",
+    "        },{\n",
+    "            \"cn_img\": base64.b64encode(image).decode('utf-8'),\n",
+    "            \"cn_stop\": 0.6,\n",
+    "            \"cn_weight\": 0.6,\n",
+    "            \"cn_type\": \"ImagePrompt\"\n",
+    "        }]\n",
+    "    }\n",
+    "result = image_prompt(params)\n",
+    "print(json.dumps(result, indent=4, ensure_ascii=False))"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    " # text to image with image prompt"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import requests\n",
+    "import json\n",
+    "import base64\n",
+    "\n",
+    "# text to image with image prompt Example\n",
+    "host = \"http://127.0.0.1:8888\"\n",
+    "image = open(\"./imgs/image_prompt-1.png\", \"rb\").read()\n",
+    "source = open(\"./imgs/image_prompt-0.jpg\", \"rb\").read()\n",
+    "def image_prompt(params: dict) -> dict:\n",
+    "    \"\"\"\n",
+    "    image prompt\n",
+    "    \"\"\"\n",
+    "    response = requests.post(\n",
+    "        url=f\"{host}/v2/generation/text-to-image-with-ip\",\n",
+    "        data=json.dumps(params),\n",
+    "        headers={\"Content-Type\": \"application/json\"})\n",
+    "    return response.json()\n",
+    "\n",
+    "params = {\n",
+    "    \"prompt\": \"A bear\",\n",
+    "    \"image_prompts\": [\n",
+    "        {\n",
+    "            \"cn_img\": base64.b64encode(source).decode('utf-8'),\n",
+    "            \"cn_stop\": 0.6,\n",
+    "            \"cn_weight\": 0.6,\n",
+    "            \"cn_type\": \"ImagePrompt\"\n",
+    "        },{\n",
+    "            \"cn_img\": base64.b64encode(image).decode('utf-8'),\n",
+    "            \"cn_stop\": 0.6,\n",
+    "            \"cn_weight\": 0.6,\n",
+    "            \"cn_type\": \"ImagePrompt\"\n",
+    "        }\n",
+    "    ]\n",
+    "}\n",
+    "result = image_prompt(params)\n",
+    "print(json.dumps(result, indent=4, ensure_ascii=False))"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# describe"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import requests\n",
+    "\n",
+    "image = open(\"./imgs/target_face.png\", \"rb\").read()\n",
+    "def describe_image(image: bytes,\n",
+    "                   params: dict = {\"type\": \"Photo\"}) -> dict:\n",
+    "    \"\"\"\n",
+    "    describe-image\n",
+    "    \"\"\"\n",
+    "    response = requests.post(\n",
+    "        url=\"http://127.0.0.1:8888/v1/tools/describe-image\",\n",
+    "        params=params,\n",
+    "        files={\n",
+    "            \"image\": image\n",
+    "        },\n",
+    "        timeout=30)\n",
+    "    return response.json()\n",
+    "\n",
+    "describe_image(image=image)"
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.10.13"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}

examples/examples_v1.py

@@ -0,0 +1,266 @@
+"""
+Examples codes for Fooocus API
+"""
+import json
+import os
+import requests
+
+
+class Config:
+    """
+    Config
+    Attributes:
+        fooocus_host (str): Fooocus API host
+        img_upscale (str): Upscale or Vary
+        inpaint_outpaint (str): Inpaint or Outpaint
+        img_prompt (str): Image Prompt
+    """
+    fooocus_host = 'http://127.0.0.1:8888'
+
+    text2image = '/v1/generation/text-to-image'
+    img_upscale = '/v1/generation/image-upscale-vary'
+    inpaint_outpaint = '/v1/generation/image-inpaint-outpaint'
+    img_prompt = '/v1/generation/image-prompt'
+
+
+def read_image(image_name: str) -> bytes:
+    """
+    Read image from file
+    Args:
+        image_name (str): Image file name
+    Returns:
+        image (bytes): Image data
+    """
+    path = os.path.join('imgs', image_name)
+    with open(path, "rb") as f:
+        image = f.read()
+        f.close()
+    return image
+
+
+class ImageList:
+    """
+    Image List
+    """
+    bear = read_image('bear.jpg')
+    image_prompt_0 = read_image('image_prompt-0.jpg')
+    image_prompt_1 = read_image('image_prompt-1.png')
+    image_prompt_2 = read_image('image_prompt-2.png')
+    image_prompt_3 = read_image('image_prompt-3.png')
+    inpaint_source = read_image('inpaint_source.jpg')
+    inpaint_mask = read_image('inpaint_mask.png')
+    source_face_f = read_image('source_face_female.png')
+    source_face_m = read_image('source_face_man.png')
+    target_face = read_image('target_face.png')
+
+
+def text2image(params: dict) -> dict:
+    """
+    Text to image
+    Args:
+        params (dict): Params
+    Returns:
+        dict: Response
+    """
+    data = json.dumps(params)
+    response = requests.post(
+        url=f"{Config.fooocus_host}{Config.text2image}",
+        data=data,
+        timeout=300)
+    return response.json()
+
+
+def upscale_vary(image: bytes, params: dict) -> dict:
+    """
+    Upscale or Vary
+    Args:
+        image (bytes): Image data
+        params (dict): Params
+    Returns:
+        dict: Response
+    """
+    response = requests.post(
+        url=f"{Config.fooocus_host}{Config.img_upscale}",
+        data=params,
+        files={
+            'input_image': image,
+        },
+        timeout=300)
+    return response.json()
+
+
+def inpaint_outpaint(
+        params: dict,
+        input_image: bytes,
+        input_mask: bytes = None) -> dict:
+    """
+    Inpaint or Outpaint
+    Args:
+        params (dict): Params
+        input_image (bytes): Image data
+        input_mask (bytes): Image mask data
+    Returns:
+        dict: Response
+    """
+    response = requests.post(
+        url=f"{Config.fooocus_host}{Config.inpaint_outpaint}",
+        data=params,
+        files={
+            'input_image': input_image,
+            'input_mask': input_mask
+        },
+        timeout=300)
+    return response.json()
+
+
+def image_prompt(
+        params: dict,
+        input_image: bytes = None,
+        input_mask: bytes = None,
+        cn_img1: bytes = None,
+        cn_img2: bytes = None,
+        cn_img3: bytes = None,
+        cn_img4: bytes = None,) -> dict:
+    """
+    Image Prompt
+    Args:
+        params (dict): Params
+        input_image (bytes): Image data
+        input_mask (bytes): Image mask data
+        cn_img1 (bytes): Image data
+        cn_img2 (bytes): Image data
+        cn_img3 (bytes): Image data
+        cn_img4 (bytes): Image data
+    Returns:
+        dict: Response
+    """
+    response = requests.post(
+        url=f"{Config.fooocus_host}{Config.img_prompt}",
+        data=params,
+        files={
+            'input_image': input_image,
+            'input_mask': input_mask,
+            'cn_img1': cn_img1,
+            'cn_img2': cn_img2,
+            'cn_img3': cn_img3,
+            'cn_img4': cn_img4
+        },
+        timeout=300)
+    return response.json()
+
+
+# ###############################################################
+# Text to image example
+# ################################################################
+
+# Text to image example
+t2i_params = {
+    "prompt": "a cat",
+    "performance_selection": "Lightning",
+    "aspect_ratios_selection": "896*1152",
+    "async_process": True
+}
+
+t2i_result = text2image(params=t2i_params)
+print(json.dumps(t2i_result))
+
+
+# ###############################################################
+# Upscale or Vary example
+# ################################################################
+
+# Upscale or Vary example
+up_params = {
+    "uov_method": "Upscale (2x)",
+    "async_process": True
+}
+
+up_result = upscale_vary(
+    image=ImageList.bear,
+    params=up_params
+)
+print(json.dumps(up_result))
+
+
+# ###############################################################
+# Inpaint or Outpaint example
+# ################################################################
+
+# Inpaint or Outpaint example
+io_params = {
+    "prompt": "a cat",
+    "outpaint_selections": "Left,Top",
+    "async_process": True
+}
+
+io_result = inpaint_outpaint(
+    params=io_params,
+    input_image=ImageList.inpaint_source,
+    input_mask=ImageList.inpaint_mask
+)
+print(json.dumps(io_result))
+
+
+# ###############################################################
+# Image prompt example
+# ################################################################
+
+# Image prompt example
+ip_params = {
+    "prompt": "a cat",
+    "image_prompts": [
+        {
+            "cn_stop": 0.6,
+            "cn_weight": 0.6,
+            "cn_type": "ImagePrompt"
+        },
+        {
+            "cn_stop": 0.6,
+            "cn_weight": 0.6,
+            "cn_type": "ImagePrompt"
+        }
+    ]
+}
+
+ip_result = image_prompt(
+    params=ip_params,
+    cn_img1=ImageList.image_prompt_0,
+    cn_img2=ImageList.image_prompt_1,
+)
+print(json.dumps(ip_result))
+
+# ###############################################################
+# Image Enhance
+# ################################################################
+
+# Image Enhance
+
+import requests
+
+url = "http://localhost:8888/v1/generation/image-enhance"
+
+# Define the file path and other form data
+file_path = "./examples/imgs/source_face_man.png"
+form_data = {
+    "enhance_checkbox": True,
+    "enhance_uov_method": "Disabled",
+    "enhance_enabled_1": True,
+    "enhance_mask_dino_prompt_1": "face",
+    "enhance_enabled_2": True,
+    "enhance_mask_dino_prompt_2": "eyes",
+}
+
+# Open the file and prepare it for the request
+with open(file_path, "rb") as f:
+    image = f.read()
+    f.close()
+
+# Send the request
+response = requests.post(
+    url,
+    files={"enhance_input_image": image},
+    data=form_data,
+    timeout=180)
+
+# Print the response content
+print(response.text)

examples/examples_v2.py

@@ -0,0 +1,288 @@
+"""
+Examples codes for Fooocus API
+"""
+import json
+import os
+import base64
+import requests
+
+
+class Config:
+    """
+    Config
+    Attributes:
+        fooocus_host (str): Fooocus API host
+        text2img_ip (str): Text to Image with IP
+        img_upscale (str): Upscale or Vary
+        inpaint_outpaint (str): Inpaint or Outpaint
+        img_prompt (str): Image Prompt
+    """
+    fooocus_host = 'http://127.0.0.1:8888'
+
+    text2img_ip = '/v2/generation/text-to-image-with-ip'
+    img_upscale = '/v2/generation/image-upscale-vary'
+    inpaint_outpaint = '/v2/generation/image-inpaint-outpaint'
+    img_prompt = '/v2/generation/image-prompt'
+
+
+def read_image(image_name: str) -> str:
+    """
+    Read image from file
+    Args:
+        image_name (str): Image file name
+    Returns:
+        str: Image base64
+    """
+    path = os.path.join('imgs', image_name)
+    with open(path, "rb") as f:
+        image = f.read()
+        f.close()
+    return base64.b64encode(image).decode('utf-8')
+
+
+class ImageList:
+    """
+    Image List
+    """
+    bear = read_image('bear.jpg')
+    image_prompt_0 = read_image('image_prompt-0.jpg')
+    image_prompt_1 = read_image('image_prompt-1.png')
+    image_prompt_2 = read_image('image_prompt-2.png')
+    image_prompt_3 = read_image('image_prompt-3.png')
+    inpaint_source = read_image('inpaint_source.jpg')
+    inpaint_mask = read_image('inpaint_mask.png')
+    source_face_f = read_image('source_face_female.png')
+    source_face_m = read_image('source_face_man.png')
+    target_face = read_image('target_face.png')
+
+
+def upscale_vary(params: dict) -> dict:
+    """
+    Upscale or Vary
+    Args:
+        params (dict): Params
+    Returns:
+        dict: Response
+    """
+    data = json.dumps(params)
+    response = requests.post(
+        url=f"{Config.fooocus_host}{Config.img_upscale}",
+        data=data,
+        timeout=300)
+    return response.json()
+
+
+def inpaint_outpaint(params: dict = None) -> dict:
+    """
+    Inpaint or Outpaint
+    Args:
+        params (dict): Params
+    Returns:
+        dict: Response
+    """
+    data = json.dumps(params)
+    response = requests.post(
+        url=f"{Config.fooocus_host}{Config.inpaint_outpaint}",
+        data=data,
+        timeout=300)
+    return response.json()
+
+
+def image_prompt(params: dict) -> dict:
+    """
+    Image Prompt
+    Args:
+        params (dict): Params
+    Returns:
+        dict: Response
+    """
+    data = json.dumps(params)
+    response = requests.post(
+        url=f"{Config.fooocus_host}{Config.img_prompt}",
+        data=data,
+        timeout=300)
+    return response.json()
+
+
+def text2image_image_prompt(params: dict) -> dict:
+    """
+    Text to image with image Prompt
+    Args:
+        params (dict): Params
+    Returns:
+        dict: Response
+    """
+    params["outpaint_selections"] = ["Left", "Right"]
+    data = json.dumps(params)
+    response = requests.post(
+        url=f"{Config.fooocus_host}{Config.text2img_ip}",
+        data=data,
+        timeout=300)
+    return response.json()
+
+
+# ################################################################
+# Upscale or Vary
+# ################################################################
+
+# Upscale (2x) example
+uov_params = {
+    "input_image": ImageList.image_prompt_0,
+    "performance_selection": "Lightning",
+    "uov_method": "Upscale (2x)",
+    "async_process": True
+}
+
+upscale_result = upscale_vary(params=uov_params)
+print(
+    json.dumps(
+        upscale_result,
+        indent=4,
+        ensure_ascii=False
+    ))
+
+# Vary (Strong) example
+uov_params['uov_method'] = 'Vary (Strong)'
+
+vary_result = upscale_vary(params=uov_params)
+print(
+    json.dumps(
+        vary_result,
+        indent=4,
+        ensure_ascii=False
+    ))
+
+
+# ################################################################
+# Inpaint or Outpaint
+# ################################################################
+
+# Inpaint outpaint example
+inpaint_params = {
+    "prompt": "a cat",  # use background prompt to remove anything what you don't want
+    "performance_selection": "Speed",  # use Lightning the quality is not good
+    "input_image": ImageList.inpaint_source,
+    "input_mask": ImageList.inpaint_mask,
+    "outpaint_selections": ["Left", "Right"],
+    "async_process": False
+}
+
+inpaint_result = inpaint_outpaint(params=inpaint_params)
+print(json.dumps(inpaint_result))
+
+
+# ################################################################
+# Image Prompt example
+# more detail for image prompt can be found:
+# https://github.com/lllyasviel/Fooocus/discussions/557
+# ################################################################
+
+# face swap example
+# This parameter comes from the default parameter of the Fooocus interface,
+# but the effect of using this parameter for face swap with Fooocus is general.
+# It is too large for the return of the original image. If necessary, try to adjust more parameters.
+face_swap_params = {
+    "performance_selection": "Speed",
+    "aspect_ratios_selection": "896*1152",
+    "image_prompts": [
+        {
+            "cn_img": ImageList.source_face_m,
+            "cn_stop": 0.5,
+            "cn_weight": 0.6,
+            "cn_type": "ImagePrompt"
+        }, {
+            "cn_img": ImageList.target_face,
+            "cn_stop": 0.9,
+            "cn_weight": 0.75,
+            "cn_type": "FaceSwap"
+        }
+    ],
+    "async_process": False
+}
+
+face_swap_result = image_prompt(params=face_swap_params)
+print(json.dumps(face_swap_result))
+
+
+# ################################################################
+# Text to image with image Prompt
+# ################################################################
+
+# Text to image with image Prompt example
+t2i_ip_params = {
+    "prompt": "a cat",
+    "performance_selection": "Speed",
+    "image_prompts": [
+        {
+            "cn_img": ImageList.image_prompt_1,
+            "cn_stop": 0.6,
+            "cn_weight": 0.8,
+            "cn_type": "ImagePrompt"
+        }, {
+            "cn_img": ImageList.image_prompt_2,
+            "cn_stop": 0.6,
+            "cn_weight": 0.6,
+            "cn_type": "ImagePrompt"
+        }
+    ],
+    "async_process": False
+}
+
+t2i_ip_result = text2image_image_prompt(params=t2i_ip_params)
+print(json.dumps(t2i_ip_result))
+
+# ################################################################
+# Image Enhance
+# ################################################################
+
+# Image Enhance
+
+import requests
+import json
+
+url = "http://localhost:8888/v2/generation/image-enhance"
+
+headers = {
+    "Content-Type": "application/json"
+}
+
+data = {
+    "enhance_input_image": "https://raw.githubusercontent.com/mrhan1993/Fooocus-API/main/examples/imgs/source_face_man.png",
+    "enhance_checkbox": True,
+    "enhance_uov_method": "Vary (Strong)",
+    "enhance_uov_processing_order": "Before First Enhancement",
+    "enhance_uov_prompt_type": "Original Prompts",
+    "enhance_ctrlnets": [
+        {
+            "enhance_enabled": True,
+            "enhance_mask_dino_prompt": "face",
+            "enhance_prompt": "",
+            "enhance_negative_prompt": "",
+            "enhance_mask_model": "sam",
+            "enhance_mask_cloth_category": "full",
+            "enhance_mask_sam_model": "vit_b",
+            "enhance_mask_text_threshold": 0.25,
+            "enhance_mask_box_threshold": 0.3,
+            "enhance_mask_sam_max_detections": 0,
+            "enhance_inpaint_disable_initial_latent": False,
+            "enhance_inpaint_engine": "v2.6",
+            "enhance_inpaint_strength": 1.0,
+            "enhance_inpaint_respective_field": 0.618,
+            "enhance_inpaint_erode_or_dilate": 0.0,
+            "enhance_mask_invert": False
+        }
+    ]
+}
+
+response = requests.post(
+    url,
+    headers=headers,
+    data=json.dumps(data),
+    timeout=180)
+
+if response.status_code == 200:
+    print("Request successful!")
+    print("Response:", response.json())
+else:
+    print("Request failed with status code:", response.status_code)
+    print("Response:", response.text)

extras/BLIP/configs/bert_config.json

@@ -0,0 +1,21 @@
+{
+  "architectures": [
+    "BertModel"
+  ],
+  "attention_probs_dropout_prob": 0.1,
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 768,
+  "initializer_range": 0.02,
+  "intermediate_size": 3072,
+  "layer_norm_eps": 1e-12,
+  "max_position_embeddings": 512,
+  "model_type": "bert",
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "pad_token_id": 0,
+  "type_vocab_size": 2,
+  "vocab_size": 30522,
+  "encoder_width": 768,
+  "add_cross_attention": true   
+}

extras/BLIP/configs/caption_coco.yaml

@@ -0,0 +1,33 @@
+image_root: '/export/share/datasets/vision/coco/images/'
+ann_root: 'annotation'
+coco_gt_root: 'annotation/coco_gt'
+
+# set pretrained as a file path or an url
+pretrained: 'https://storage.googleapis.com/sfr-vision-language-research/BLIP/models/model_base_caption_capfilt_large.pth'
+
+# size of vit model; base or large
+vit: 'base'
+vit_grad_ckpt: False
+vit_ckpt_layer: 0
+batch_size: 32
+init_lr: 1e-5
+
+# vit: 'large'
+# vit_grad_ckpt: True
+# vit_ckpt_layer: 5
+# batch_size: 16
+# init_lr: 2e-6
+
+image_size: 384
+
+# generation configs
+max_length: 20  
+min_length: 5
+num_beams: 3
+prompt: 'a picture of '
+
+# optimizer
+weight_decay: 0.05
+min_lr: 0
+max_epoch: 5
+

extras/BLIP/configs/med_config.json

@@ -0,0 +1,21 @@
+{
+  "architectures": [
+    "BertModel"
+  ],
+  "attention_probs_dropout_prob": 0.1,
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 768,
+  "initializer_range": 0.02,
+  "intermediate_size": 3072,
+  "layer_norm_eps": 1e-12,
+  "max_position_embeddings": 512,
+  "model_type": "bert",
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "pad_token_id": 0,
+  "type_vocab_size": 2,
+  "vocab_size": 30524,
+  "encoder_width": 768,
+  "add_cross_attention": true   
+}

extras/BLIP/configs/nlvr.yaml

@@ -0,0 +1,21 @@
+image_root: '/export/share/datasets/vision/NLVR2/' 
+ann_root: 'annotation'
+
+# set pretrained as a file path or an url
+pretrained: 'https://storage.googleapis.com/sfr-vision-language-research/BLIP/models/model_base_nlvr.pth'
+
+#size of vit model; base or large
+vit: 'base'
+batch_size_train: 16 
+batch_size_test: 64 
+vit_grad_ckpt: False
+vit_ckpt_layer: 0
+max_epoch: 15
+
+image_size: 384
+
+# optimizer
+weight_decay: 0.05
+init_lr: 3e-5
+min_lr: 0
+

extras/BLIP/configs/nocaps.yaml

@@ -0,0 +1,15 @@
+image_root: '/export/share/datasets/vision/nocaps/'
+ann_root: 'annotation'
+
+# set pretrained as a file path or an url
+pretrained: 'https://storage.googleapis.com/sfr-vision-language-research/BLIP/models/model_base_caption_capfilt_large.pth'
+
+vit: 'base'
+batch_size: 32
+
+image_size: 384
+
+max_length: 20
+min_length: 5
+num_beams: 3
+prompt: 'a picture of '

extras/BLIP/configs/pretrain.yaml

@@ -0,0 +1,27 @@
+train_file: ['/export/share/junnan-li/VL_pretrain/annotation/coco_karpathy_train.json',
+             '/export/share/junnan-li/VL_pretrain/annotation/vg_caption.json',
+             ]
+laion_path: ''   
+
+# size of vit model; base or large
+vit: 'base'
+vit_grad_ckpt: False
+vit_ckpt_layer: 0
+
+image_size: 224
+batch_size: 75
+
+queue_size: 57600
+alpha: 0.4
+
+# optimizer
+weight_decay: 0.05
+init_lr: 3e-4
+min_lr: 1e-6
+warmup_lr: 1e-6
+lr_decay_rate: 0.9
+max_epoch: 20
+warmup_steps: 3000
+
+
+

extras/BLIP/configs/retrieval_coco.yaml

@@ -0,0 +1,34 @@
+image_root: '/export/share/datasets/vision/coco/images/'
+ann_root: 'annotation'
+dataset: 'coco'
+
+# set pretrained as a file path or an url
+pretrained: 'https://storage.googleapis.com/sfr-vision-language-research/BLIP/models/model_base_retrieval_coco.pth'
+
+# size of vit model; base or large
+
+vit: 'base'
+batch_size_train: 32
+batch_size_test: 64
+vit_grad_ckpt: True
+vit_ckpt_layer: 4
+init_lr: 1e-5
+
+# vit: 'large'
+# batch_size_train: 16
+# batch_size_test: 32
+# vit_grad_ckpt: True
+# vit_ckpt_layer: 12
+# init_lr: 5e-6
+
+image_size: 384
+queue_size: 57600
+alpha: 0.4
+k_test: 256
+negative_all_rank: True
+
+# optimizer
+weight_decay: 0.05
+min_lr: 0
+max_epoch: 6
+

extras/BLIP/configs/retrieval_flickr.yaml

@@ -0,0 +1,34 @@
+image_root: '/export/share/datasets/vision/flickr30k/'
+ann_root: 'annotation'
+dataset: 'flickr'
+
+# set pretrained as a file path or an url
+pretrained: 'https://storage.googleapis.com/sfr-vision-language-research/BLIP/models/model_base_retrieval_flickr.pth'
+
+# size of vit model; base or large
+
+vit: 'base'
+batch_size_train: 32
+batch_size_test: 64
+vit_grad_ckpt: True
+vit_ckpt_layer: 4
+init_lr: 1e-5
+
+# vit: 'large'
+# batch_size_train: 16
+# batch_size_test: 32
+# vit_grad_ckpt: True
+# vit_ckpt_layer: 10
+# init_lr: 5e-6
+
+image_size: 384
+queue_size: 57600
+alpha: 0.4
+k_test: 128
+negative_all_rank: False
+
+# optimizer
+weight_decay: 0.05
+min_lr: 0
+max_epoch: 6
+

extras/BLIP/configs/retrieval_msrvtt.yaml

@@ -0,0 +1,12 @@
+video_root: '/export/share/dongxuli/data/msrvtt_retrieval/videos'
+ann_root: 'annotation'
+
+# set pretrained as a file path or an url
+pretrained: 'https://storage.googleapis.com/sfr-vision-language-research/BLIP/models/model_base_retrieval_coco.pth'
+
+# size of vit model; base or large
+vit: 'base'
+batch_size: 64
+k_test: 128
+image_size: 384
+num_frm_test: 8

extras/BLIP/configs/vqa.yaml

@@ -0,0 +1,25 @@
+vqa_root: '/export/share/datasets/vision/VQA/Images/mscoco/' #followed by train2014/
+vg_root: '/export/share/datasets/vision/visual-genome/'  #followed by image/
+train_files: ['vqa_train','vqa_val','vg_qa']
+ann_root: 'annotation'
+
+# set pretrained as a file path or an url
+pretrained: 'https://storage.googleapis.com/sfr-vision-language-research/BLIP/models/model_base_vqa_capfilt_large.pth'
+
+# size of vit model; base or large
+vit: 'base'
+batch_size_train: 16 
+batch_size_test: 32 
+vit_grad_ckpt: False
+vit_ckpt_layer: 0
+init_lr: 2e-5
+
+image_size: 480
+
+k_test: 128
+inference: 'rank'
+
+# optimizer
+weight_decay: 0.05
+min_lr: 0
+max_epoch: 10

extras/BLIP/models/bert_tokenizer/config.json

@@ -0,0 +1,23 @@
+{
+  "architectures": [
+    "BertForMaskedLM"
+  ],
+  "attention_probs_dropout_prob": 0.1,
+  "gradient_checkpointing": false,
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 768,
+  "initializer_range": 0.02,
+  "intermediate_size": 3072,
+  "layer_norm_eps": 1e-12,
+  "max_position_embeddings": 512,
+  "model_type": "bert",
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "pad_token_id": 0,
+  "position_embedding_type": "absolute",
+  "transformers_version": "4.6.0.dev0",
+  "type_vocab_size": 2,
+  "use_cache": true,
+  "vocab_size": 30522
+}

extras/BLIP/models/bert_tokenizer/tokenizer_config.json

@@ -0,0 +1,3 @@
+{
+  "do_lower_case": true
+}

extras/BLIP/models/blip.py

@@ -0,0 +1,239 @@
+'''
+ * Copyright (c) 2022, salesforce.com, inc.
+ * All rights reserved.
+ * SPDX-License-Identifier: BSD-3-Clause
+ * For full license text, see LICENSE.txt file in the repo root or https://opensource.org/licenses/BSD-3-Clause
+ * By Junnan Li
+'''
+import warnings
+warnings.filterwarnings("ignore")
+
+from extras.BLIP.models.vit import VisionTransformer, interpolate_pos_embed
+from extras.BLIP.models.med import BertConfig, BertModel, BertLMHeadModel
+from transformers import BertTokenizer
+
+import torch
+from torch import nn
+import torch.nn.functional as F
+
+import os
+from urllib.parse import urlparse
+from timm.models.hub import download_cached_file
+
+class BLIP_Base(nn.Module):
+    def __init__(self,                 
+                 med_config = 'configs/med_config.json',  
+                 image_size = 224,
+                 vit = 'base',
+                 vit_grad_ckpt = False,
+                 vit_ckpt_layer = 0,                 
+                 ):
+        """
+        Args:
+            med_config (str): path for the mixture of encoder-decoder model's configuration file
+            image_size (int): input image size
+            vit (str): model size of vision transformer
+        """               
+        super().__init__()
+        
+        self.visual_encoder, vision_width = create_vit(vit,image_size, vit_grad_ckpt, vit_ckpt_layer)
+        self.tokenizer = init_tokenizer()   
+        med_config = BertConfig.from_json_file(med_config)
+        med_config.encoder_width = vision_width
+        self.text_encoder = BertModel(config=med_config, add_pooling_layer=False)  
+
+        
+    def forward(self, image, caption, mode):
+        
+        assert mode in ['image', 'text', 'multimodal'], "mode parameter must be image, text, or multimodal"
+        text = self.tokenizer(caption, return_tensors="pt").to(image.device) 
+        
+        if mode=='image':    
+            # return image features
+            image_embeds = self.visual_encoder(image)             
+            return image_embeds
+        
+        elif mode=='text':
+            # return text features
+            text_output = self.text_encoder(text.input_ids, attention_mask = text.attention_mask,                      
+                                            return_dict = True, mode = 'text')  
+            return text_output.last_hidden_state
+        
+        elif mode=='multimodal':
+            # return multimodel features
+            image_embeds = self.visual_encoder(image)    
+            image_atts = torch.ones(image_embeds.size()[:-1],dtype=torch.long).to(image.device)      
+            
+            text.input_ids[:,0] = self.tokenizer.enc_token_id
+            output = self.text_encoder(text.input_ids,
+                                       attention_mask = text.attention_mask,
+                                       encoder_hidden_states = image_embeds,
+                                       encoder_attention_mask = image_atts,      
+                                       return_dict = True,
+                                      )              
+            return output.last_hidden_state
+        
+        
+        
+class BLIP_Decoder(nn.Module):
+    def __init__(self,                 
+                 med_config = 'configs/med_config.json',  
+                 image_size = 384,
+                 vit = 'base',
+                 vit_grad_ckpt = False,
+                 vit_ckpt_layer = 0,
+                 prompt = 'a picture of ',
+                 ):
+        """
+        Args:
+            med_config (str): path for the mixture of encoder-decoder model's configuration file
+            image_size (int): input image size
+            vit (str): model size of vision transformer
+        """            
+        super().__init__()
+        
+        self.visual_encoder, vision_width = create_vit(vit,image_size, vit_grad_ckpt, vit_ckpt_layer)
+        self.tokenizer = init_tokenizer()   
+        med_config = BertConfig.from_json_file(med_config)
+        med_config.encoder_width = vision_width
+        self.text_decoder = BertLMHeadModel(config=med_config)    
+        
+        self.prompt = prompt
+        self.prompt_length = len(self.tokenizer(self.prompt).input_ids)-1
+
+        
+    def forward(self, image, caption):
+        
+        image_embeds = self.visual_encoder(image) 
+        image_atts = torch.ones(image_embeds.size()[:-1],dtype=torch.long).to(image.device)
+        
+        text = self.tokenizer(caption, padding='longest', truncation=True, max_length=40, return_tensors="pt").to(image.device) 
+        
+        text.input_ids[:,0] = self.tokenizer.bos_token_id
+        
+        decoder_targets = text.input_ids.masked_fill(text.input_ids == self.tokenizer.pad_token_id, -100)         
+        decoder_targets[:,:self.prompt_length] = -100
+     
+        decoder_output = self.text_decoder(text.input_ids, 
+                                           attention_mask = text.attention_mask, 
+                                           encoder_hidden_states = image_embeds,
+                                           encoder_attention_mask = image_atts,                  
+                                           labels = decoder_targets,
+                                           return_dict = True,   
+                                          )   
+        loss_lm = decoder_output.loss
+        
+        return loss_lm
+        
+    def generate(self, image, sample=False, num_beams=3, max_length=30, min_length=10, top_p=0.9, repetition_penalty=1.0):
+        image_embeds = self.visual_encoder(image)
+
+        if not sample:
+            image_embeds = image_embeds.repeat_interleave(num_beams,dim=0)
+            
+        image_atts = torch.ones(image_embeds.size()[:-1],dtype=torch.long).to(image.device)
+        model_kwargs = {"encoder_hidden_states": image_embeds, "encoder_attention_mask":image_atts}
+        
+        prompt = [self.prompt] * image.size(0)
+        input_ids = self.tokenizer(prompt, return_tensors="pt").input_ids.to(image.device) 
+        input_ids[:,0] = self.tokenizer.bos_token_id
+        input_ids = input_ids[:, :-1] 
+
+        if sample:
+            #nucleus sampling
+            outputs = self.text_decoder.generate(input_ids=input_ids,
+                                                  max_length=max_length,
+                                                  min_length=min_length,
+                                                  do_sample=True,
+                                                  top_p=top_p,
+                                                  num_return_sequences=1,
+                                                  eos_token_id=self.tokenizer.sep_token_id,
+                                                  pad_token_id=self.tokenizer.pad_token_id, 
+                                                  repetition_penalty=1.1,                                            
+                                                  **model_kwargs)
+        else:
+            #beam search
+            outputs = self.text_decoder.generate(input_ids=input_ids,
+                                                  max_length=max_length,
+                                                  min_length=min_length,
+                                                  num_beams=num_beams,
+                                                  eos_token_id=self.tokenizer.sep_token_id,
+                                                  pad_token_id=self.tokenizer.pad_token_id,     
+                                                  repetition_penalty=repetition_penalty,
+                                                  **model_kwargs)            
+            
+        captions = []    
+        for output in outputs:
+            caption = self.tokenizer.decode(output, skip_special_tokens=True)    
+            captions.append(caption[len(self.prompt):])
+        return captions
+    
+
+def blip_decoder(pretrained='',**kwargs):
+    model = BLIP_Decoder(**kwargs)
+    if pretrained:
+        model,msg = load_checkpoint(model,pretrained)
+        assert(len(msg.missing_keys)==0)
+    return model    
+    
+def blip_feature_extractor(pretrained='',**kwargs):
+    model = BLIP_Base(**kwargs)
+    if pretrained:
+        model,msg = load_checkpoint(model,pretrained)
+        assert(len(msg.missing_keys)==0)
+    return model        
+
+def init_tokenizer():
+    tokenizer_path = os.path.join(os.path.dirname(os.path.realpath(__file__)), "bert_tokenizer")
+    tokenizer = BertTokenizer.from_pretrained(tokenizer_path)
+    tokenizer.add_special_tokens({'bos_token':'[DEC]'})
+    tokenizer.add_special_tokens({'additional_special_tokens':['[ENC]']})       
+    tokenizer.enc_token_id = tokenizer.additional_special_tokens_ids[0]  
+    return tokenizer
+
+
+def create_vit(vit, image_size, use_grad_checkpointing=False, ckpt_layer=0, drop_path_rate=0):
+        
+    assert vit in ['base', 'large'], "vit parameter must be base or large"
+    if vit=='base':
+        vision_width = 768
+        visual_encoder = VisionTransformer(img_size=image_size, patch_size=16, embed_dim=vision_width, depth=12, 
+                                           num_heads=12, use_grad_checkpointing=use_grad_checkpointing, ckpt_layer=ckpt_layer,
+                                           drop_path_rate=0 or drop_path_rate
+                                          )   
+    elif vit=='large':
+        vision_width = 1024
+        visual_encoder = VisionTransformer(img_size=image_size, patch_size=16, embed_dim=vision_width, depth=24, 
+                                           num_heads=16, use_grad_checkpointing=use_grad_checkpointing, ckpt_layer=ckpt_layer,
+                                           drop_path_rate=0.1 or drop_path_rate
+                                          )   
+    return visual_encoder, vision_width
+
+def is_url(url_or_filename):
+    parsed = urlparse(url_or_filename)
+    return parsed.scheme in ("http", "https")
+
+def load_checkpoint(model,url_or_filename):
+    if is_url(url_or_filename):
+        cached_file = download_cached_file(url_or_filename, check_hash=False, progress=True)
+        checkpoint = torch.load(cached_file, map_location='cpu', weights_only=True) 
+    elif os.path.isfile(url_or_filename):        
+        checkpoint = torch.load(url_or_filename, map_location='cpu', weights_only=True) 
+    else:
+        raise RuntimeError('checkpoint url or path is invalid')
+        
+    state_dict = checkpoint['model']
+    
+    state_dict['visual_encoder.pos_embed'] = interpolate_pos_embed(state_dict['visual_encoder.pos_embed'],model.visual_encoder) 
+    if 'visual_encoder_m.pos_embed' in model.state_dict().keys():
+        state_dict['visual_encoder_m.pos_embed'] = interpolate_pos_embed(state_dict['visual_encoder_m.pos_embed'],
+                                                                         model.visual_encoder_m)    
+    for key in model.state_dict().keys():
+        if key in state_dict.keys():
+            if state_dict[key].shape!=model.state_dict()[key].shape:
+                del state_dict[key]
+    
+    msg = model.load_state_dict(state_dict,strict=False)
+    print('load checkpoint from %s'%url_or_filename)  
+    return model,msg
+    

extras/BLIP/models/blip_itm.py

@@ -0,0 +1,76 @@
+from extras.BLIP.models.med import BertConfig, BertModel
+from transformers import BertTokenizer
+
+import torch
+from torch import nn
+import torch.nn.functional as F
+
+from extras.BLIP.models.blip import create_vit, init_tokenizer, load_checkpoint
+
+class BLIP_ITM(nn.Module):
+    def __init__(self,                 
+                 med_config = 'configs/med_config.json',  
+                 image_size = 384,
+                 vit = 'base',
+                 vit_grad_ckpt = False,
+                 vit_ckpt_layer = 0,                      
+                 embed_dim = 256,     
+                 ):
+        """
+        Args:
+            med_config (str): path for the mixture of encoder-decoder model's configuration file
+            image_size (int): input image size
+            vit (str): model size of vision transformer
+        """               
+        super().__init__()
+        
+        self.visual_encoder, vision_width = create_vit(vit,image_size, vit_grad_ckpt, vit_ckpt_layer)
+        self.tokenizer = init_tokenizer()   
+        med_config = BertConfig.from_json_file(med_config)
+        med_config.encoder_width = vision_width
+        self.text_encoder = BertModel(config=med_config, add_pooling_layer=False)          
+
+        text_width = self.text_encoder.config.hidden_size
+        
+        self.vision_proj = nn.Linear(vision_width, embed_dim)
+        self.text_proj = nn.Linear(text_width, embed_dim)
+
+        self.itm_head = nn.Linear(text_width, 2) 
+        
+        
+    def forward(self, image, caption, match_head='itm'):
+
+        image_embeds = self.visual_encoder(image) 
+        image_atts = torch.ones(image_embeds.size()[:-1],dtype=torch.long).to(image.device)        
+      
+        text = self.tokenizer(caption, padding='max_length', truncation=True, max_length=35, 
+                              return_tensors="pt").to(image.device) 
+
+                 
+        if match_head=='itm':
+            output = self.text_encoder(text.input_ids,
+                                       attention_mask = text.attention_mask,
+                                       encoder_hidden_states = image_embeds,
+                                       encoder_attention_mask = image_atts,      
+                                       return_dict = True,
+                                      )
+            itm_output = self.itm_head(output.last_hidden_state[:,0,:])     
+            return itm_output
+            
+        elif match_head=='itc':
+            text_output = self.text_encoder(text.input_ids, attention_mask = text.attention_mask,                      
+                                            return_dict = True, mode = 'text')                     
+            image_feat = F.normalize(self.vision_proj(image_embeds[:,0,:]),dim=-1)   
+            text_feat = F.normalize(self.text_proj(text_output.last_hidden_state[:,0,:]),dim=-1)    
+            
+            sim = image_feat @ text_feat.t()
+            return sim
+        
+        
+def blip_itm(pretrained='',**kwargs):
+    model = BLIP_ITM(**kwargs)
+    if pretrained:
+        model,msg = load_checkpoint(model,pretrained)
+        assert(len(msg.missing_keys)==0)
+    return model         
+            

extras/BLIP/models/blip_nlvr.py

@@ -0,0 +1,105 @@
+from extras.BLIP.models.med import BertConfig
+from extras.BLIP.models.nlvr_encoder import BertModel
+from extras.BLIP.models.vit import interpolate_pos_embed
+from extras.BLIP.models.blip import create_vit, init_tokenizer, is_url
+
+from timm.models.hub import download_cached_file
+
+import torch
+from torch import nn
+import torch.nn.functional as F
+from transformers import BertTokenizer
+import numpy as np
+import os
+
+
+class BLIP_NLVR(nn.Module):
+    def __init__(self,                 
+                 med_config = 'configs/med_config.json',  
+                 image_size = 480,
+                 vit = 'base',
+                 vit_grad_ckpt = False,
+                 vit_ckpt_layer = 0,                   
+                 ):
+        """
+        Args:
+            med_config (str): path for the mixture of encoder-decoder model's configuration file
+            image_size (int): input image size
+            vit (str): model size of vision transformer
+        """               
+        super().__init__()
+        
+        self.visual_encoder, vision_width = create_vit(vit,image_size, vit_grad_ckpt, vit_ckpt_layer, drop_path_rate=0.1)
+        self.tokenizer = init_tokenizer()   
+        med_config = BertConfig.from_json_file(med_config)
+        med_config.encoder_width = vision_width
+        self.text_encoder = BertModel(config=med_config, add_pooling_layer=False) 
+                    
+        self.cls_head = nn.Sequential(
+                  nn.Linear(self.text_encoder.config.hidden_size, self.text_encoder.config.hidden_size),
+                  nn.ReLU(),
+                  nn.Linear(self.text_encoder.config.hidden_size, 2)
+                )  
+
+    def forward(self, image, text, targets, train=True):
+        
+        image_embeds = self.visual_encoder(image) 
+        image_atts = torch.ones(image_embeds.size()[:-1],dtype=torch.long).to(image.device)        
+        image0_embeds, image1_embeds = torch.split(image_embeds,targets.size(0))     
+
+        text = self.tokenizer(text, padding='longest', return_tensors="pt").to(image.device) 
+        text.input_ids[:,0] = self.tokenizer.enc_token_id        
+
+        output = self.text_encoder(text.input_ids, 
+                                   attention_mask = text.attention_mask, 
+                                   encoder_hidden_states = [image0_embeds,image1_embeds],
+                                   encoder_attention_mask = [image_atts[:image0_embeds.size(0)],
+                                                             image_atts[image0_embeds.size(0):]],        
+                                   return_dict = True,
+                                  )  
+        hidden_state = output.last_hidden_state[:,0,:]        
+        prediction = self.cls_head(hidden_state)
+
+        if train:            
+            loss = F.cross_entropy(prediction, targets)   
+            return loss
+        else:
+            return prediction
+    
+def blip_nlvr(pretrained='',**kwargs):
+    model = BLIP_NLVR(**kwargs)
+    if pretrained:
+        model,msg = load_checkpoint(model,pretrained)
+        print("missing keys:")
+        print(msg.missing_keys)
+    return model  
+
+        
+def load_checkpoint(model,url_or_filename):
+    if is_url(url_or_filename):
+        cached_file = download_cached_file(url_or_filename, check_hash=False, progress=True)
+        checkpoint = torch.load(cached_file, map_location='cpu', weights_only=True) 
+    elif os.path.isfile(url_or_filename):        
+        checkpoint = torch.load(url_or_filename, map_location='cpu', weights_only=True) 
+    else:
+        raise RuntimeError('checkpoint url or path is invalid')
+    state_dict = checkpoint['model']
+    
+    state_dict['visual_encoder.pos_embed'] = interpolate_pos_embed(state_dict['visual_encoder.pos_embed'],model.visual_encoder) 
+    
+    for key in list(state_dict.keys()):
+        if 'crossattention.self.' in key:
+            new_key0 = key.replace('self','self0')
+            new_key1 = key.replace('self','self1')
+            state_dict[new_key0] = state_dict[key]
+            state_dict[new_key1] = state_dict[key]
+        elif 'crossattention.output.dense.' in key:
+            new_key0 = key.replace('dense','dense0')
+            new_key1 = key.replace('dense','dense1')
+            state_dict[new_key0] = state_dict[key]
+            state_dict[new_key1] = state_dict[key]  
+                
+    msg = model.load_state_dict(state_dict,strict=False)
+    print('load checkpoint from %s'%url_or_filename)  
+    return model,msg
+            

extras/BLIP/models/blip_pretrain.py

@@ -0,0 +1,339 @@
+'''
+ * Copyright (c) 2022, salesforce.com, inc.
+ * All rights reserved.
+ * SPDX-License-Identifier: BSD-3-Clause
+ * For full license text, see LICENSE.txt file in the repo root or https://opensource.org/licenses/BSD-3-Clause
+ * By Junnan Li
+'''
+from extras.BLIP.models.med import BertConfig, BertModel, BertLMHeadModel
+from transformers import BertTokenizer
+import transformers
+transformers.logging.set_verbosity_error()
+
+import torch
+from torch import nn
+import torch.nn.functional as F
+
+from extras.BLIP.models.blip import create_vit, init_tokenizer, load_checkpoint
+
+class BLIP_Pretrain(nn.Module):
+    def __init__(self,                 
+                 med_config = 'configs/bert_config.json',  
+                 image_size = 224,
+                 vit = 'base',
+                 vit_grad_ckpt = False,
+                 vit_ckpt_layer = 0,                    
+                 embed_dim = 256,     
+                 queue_size = 57600,
+                 momentum = 0.995,
+                 ):
+        """
+        Args:
+            med_config (str): path for the mixture of encoder-decoder model's configuration file
+            image_size (int): input image size
+            vit (str): model size of vision transformer
+        """               
+        super().__init__()
+        
+        self.visual_encoder, vision_width = create_vit(vit,image_size, vit_grad_ckpt, vit_ckpt_layer, 0)
+        
+        if vit=='base':
+            checkpoint = torch.hub.load_state_dict_from_url(
+                url="https://dl.fbaipublicfiles.com/deit/deit_base_patch16_224-b5f2ef4d.pth",
+                map_location="cpu", check_hash=True)
+            state_dict = checkpoint["model"]     
+            msg = self.visual_encoder.load_state_dict(state_dict,strict=False)
+        elif vit=='large':
+            from timm.models.helpers import load_custom_pretrained
+            from timm.models.vision_transformer import default_cfgs
+            load_custom_pretrained(self.visual_encoder,default_cfgs['vit_large_patch16_224_in21k'])        
+               
+        self.tokenizer = init_tokenizer()   
+        encoder_config = BertConfig.from_json_file(med_config)
+        encoder_config.encoder_width = vision_width
+        self.text_encoder = BertModel.from_pretrained('bert-base-uncased',config=encoder_config, add_pooling_layer=False)
+        self.text_encoder.resize_token_embeddings(len(self.tokenizer)) 
+
+        text_width = self.text_encoder.config.hidden_size
+        
+        self.vision_proj = nn.Linear(vision_width, embed_dim)
+        self.text_proj = nn.Linear(text_width, embed_dim)
+
+        self.itm_head = nn.Linear(text_width, 2) 
+        
+        # create momentum encoders  
+        self.visual_encoder_m, vision_width = create_vit(vit,image_size)              
+        self.vision_proj_m = nn.Linear(vision_width, embed_dim)
+        self.text_encoder_m = BertModel(config=encoder_config, add_pooling_layer=False)      
+        self.text_proj_m = nn.Linear(text_width, embed_dim)
+        
+        self.model_pairs = [[self.visual_encoder,self.visual_encoder_m],
+                            [self.vision_proj,self.vision_proj_m],
+                            [self.text_encoder,self.text_encoder_m],
+                            [self.text_proj,self.text_proj_m],
+                           ]       
+        self.copy_params()
+
+        # create the queue
+        self.register_buffer("image_queue", torch.randn(embed_dim, queue_size))
+        self.register_buffer("text_queue", torch.randn(embed_dim, queue_size))
+        self.register_buffer("queue_ptr", torch.zeros(1, dtype=torch.long))  
+
+        self.image_queue = nn.functional.normalize(self.image_queue, dim=0)
+        self.text_queue = nn.functional.normalize(self.text_queue, dim=0)
+        
+        self.queue_size = queue_size
+        self.momentum = momentum
+        self.temp = nn.Parameter(0.07*torch.ones([]))   
+        
+        # create the decoder
+        decoder_config = BertConfig.from_json_file(med_config)
+        decoder_config.encoder_width = vision_width        
+        self.text_decoder = BertLMHeadModel.from_pretrained('bert-base-uncased',config=decoder_config)    
+        self.text_decoder.resize_token_embeddings(len(self.tokenizer)) 
+        tie_encoder_decoder_weights(self.text_encoder,self.text_decoder.bert,'','/attention')
+        
+        
+    def forward(self, image, caption, alpha):
+        with torch.no_grad():
+            self.temp.clamp_(0.001,0.5)
+        
+        image_embeds = self.visual_encoder(image) 
+        image_atts = torch.ones(image_embeds.size()[:-1],dtype=torch.long).to(image.device)        
+        image_feat = F.normalize(self.vision_proj(image_embeds[:,0,:]),dim=-1)          
+        
+        text = self.tokenizer(caption, padding='max_length', truncation=True, max_length=30, 
+                              return_tensors="pt").to(image.device)  
+        text_output = self.text_encoder(text.input_ids, attention_mask = text.attention_mask,                      
+                                        return_dict = True, mode = 'text')            
+        text_feat = F.normalize(self.text_proj(text_output.last_hidden_state[:,0,:]),dim=-1)                 
+             
+        # get momentum features
+        with torch.no_grad():
+            self._momentum_update()
+            image_embeds_m = self.visual_encoder_m(image) 
+            image_feat_m = F.normalize(self.vision_proj_m(image_embeds_m[:,0,:]),dim=-1)  
+            image_feat_all = torch.cat([image_feat_m.t(),self.image_queue.clone().detach()],dim=1)                   
+            
+            text_output_m = self.text_encoder_m(text.input_ids, attention_mask = text.attention_mask,                      
+                                                return_dict = True, mode = 'text')    
+            text_feat_m = F.normalize(self.text_proj_m(text_output_m.last_hidden_state[:,0,:]),dim=-1) 
+            text_feat_all = torch.cat([text_feat_m.t(),self.text_queue.clone().detach()],dim=1)
+
+            sim_i2t_m = image_feat_m @ text_feat_all / self.temp  
+            sim_t2i_m = text_feat_m @ image_feat_all / self.temp 
+
+            sim_targets = torch.zeros(sim_i2t_m.size()).to(image.device)
+            sim_targets.fill_diagonal_(1)          
+
+            sim_i2t_targets = alpha * F.softmax(sim_i2t_m, dim=1) + (1 - alpha) * sim_targets
+            sim_t2i_targets = alpha * F.softmax(sim_t2i_m, dim=1) + (1 - alpha) * sim_targets        
+
+        sim_i2t = image_feat @ text_feat_all / self.temp
+        sim_t2i = text_feat @ image_feat_all / self.temp
+                             
+        loss_i2t = -torch.sum(F.log_softmax(sim_i2t, dim=1)*sim_i2t_targets,dim=1).mean()
+        loss_t2i = -torch.sum(F.log_softmax(sim_t2i, dim=1)*sim_t2i_targets,dim=1).mean() 
+
+        loss_ita = (loss_i2t+loss_t2i)/2
+
+        self._dequeue_and_enqueue(image_feat_m, text_feat_m)        
+
+        ###============== Image-text Matching ===================###
+        encoder_input_ids = text.input_ids.clone()
+        encoder_input_ids[:,0] = self.tokenizer.enc_token_id
+        
+        # forward the positve image-text pair
+        bs = image.size(0)
+        output_pos = self.text_encoder(encoder_input_ids,
+                                       attention_mask = text.attention_mask,
+                                       encoder_hidden_states = image_embeds,
+                                       encoder_attention_mask = image_atts,      
+                                       return_dict = True,
+                                      )            
+        with torch.no_grad():       
+            weights_t2i = F.softmax(sim_t2i[:,:bs],dim=1)+1e-4 
+            weights_t2i.fill_diagonal_(0)            
+            weights_i2t = F.softmax(sim_i2t[:,:bs],dim=1)+1e-4  
+            weights_i2t.fill_diagonal_(0)   
+            
+        # select a negative image for each text
+        image_embeds_neg = []    
+        for b in range(bs):
+            neg_idx = torch.multinomial(weights_t2i[b], 1).item()
+            image_embeds_neg.append(image_embeds[neg_idx])
+        image_embeds_neg = torch.stack(image_embeds_neg,dim=0)   
+
+        # select a negative text for each image
+        text_ids_neg = []
+        text_atts_neg = []
+        for b in range(bs):
+            neg_idx = torch.multinomial(weights_i2t[b], 1).item()
+            text_ids_neg.append(encoder_input_ids[neg_idx])
+            text_atts_neg.append(text.attention_mask[neg_idx])
+
+        text_ids_neg = torch.stack(text_ids_neg,dim=0)   
+        text_atts_neg = torch.stack(text_atts_neg,dim=0)      
+
+        text_ids_all = torch.cat([encoder_input_ids, text_ids_neg],dim=0)     
+        text_atts_all = torch.cat([text.attention_mask, text_atts_neg],dim=0)     
+
+        image_embeds_all = torch.cat([image_embeds_neg,image_embeds],dim=0)
+        image_atts_all = torch.cat([image_atts,image_atts],dim=0)
+
+        output_neg = self.text_encoder(text_ids_all,
+                                       attention_mask = text_atts_all,
+                                       encoder_hidden_states = image_embeds_all,
+                                       encoder_attention_mask = image_atts_all,      
+                                       return_dict = True,
+                                      )                            
+
+        vl_embeddings = torch.cat([output_pos.last_hidden_state[:,0,:], output_neg.last_hidden_state[:,0,:]],dim=0)
+        vl_output = self.itm_head(vl_embeddings)            
+
+        itm_labels = torch.cat([torch.ones(bs,dtype=torch.long),torch.zeros(2*bs,dtype=torch.long)],
+                               dim=0).to(image.device)
+        loss_itm = F.cross_entropy(vl_output, itm_labels)  
+        
+        ##================= LM ========================##     
+        decoder_input_ids = text.input_ids.clone()      
+        decoder_input_ids[:,0] = self.tokenizer.bos_token_id
+        decoder_targets = decoder_input_ids.masked_fill(decoder_input_ids == self.tokenizer.pad_token_id, -100) 
+
+        decoder_output = self.text_decoder(decoder_input_ids, 
+                                           attention_mask = text.attention_mask, 
+                                           encoder_hidden_states = image_embeds,
+                                           encoder_attention_mask = image_atts,                  
+                                           labels = decoder_targets,
+                                           return_dict = True,   
+                                          )   
+          
+        loss_lm = decoder_output.loss                
+        return loss_ita, loss_itm, loss_lm
+ 
+
+
+    @torch.no_grad()    
+    def copy_params(self):
+        for model_pair in self.model_pairs:           
+            for param, param_m in zip(model_pair[0].parameters(), model_pair[1].parameters()):
+                param_m.data.copy_(param.data)  # initialize
+                param_m.requires_grad = False  # not update by gradient    
+
+            
+    @torch.no_grad()        
+    def _momentum_update(self):
+        for model_pair in self.model_pairs:           
+            for param, param_m in zip(model_pair[0].parameters(), model_pair[1].parameters()):
+                param_m.data = param_m.data * self.momentum + param.data * (1. - self.momentum)
+
+                        
+    @torch.no_grad()
+    def _dequeue_and_enqueue(self, image_feat, text_feat):
+        # gather keys before updating queue
+        image_feats = concat_all_gather(image_feat)
+        text_feats = concat_all_gather(text_feat)
+
+        batch_size = image_feats.shape[0]
+
+        ptr = int(self.queue_ptr)
+        assert self.queue_size % batch_size == 0  # for simplicity
+
+        # replace the keys at ptr (dequeue and enqueue)
+        self.image_queue[:, ptr:ptr + batch_size] = image_feats.T
+        self.text_queue[:, ptr:ptr + batch_size] = text_feats.T
+        ptr = (ptr + batch_size) % self.queue_size  # move pointer
+
+        self.queue_ptr[0] = ptr 
+
+
+def blip_pretrain(**kwargs):
+    model = BLIP_Pretrain(**kwargs)
+    return model 
+
+
+@torch.no_grad()
+def concat_all_gather(tensor):
+    """
+    Performs all_gather operation on the provided tensors.
+    *** Warning ***: torch.distributed.all_gather has no gradient.
+    """
+    tensors_gather = [torch.ones_like(tensor)
+        for _ in range(torch.distributed.get_world_size())]
+    torch.distributed.all_gather(tensors_gather, tensor, async_op=False)
+
+    output = torch.cat(tensors_gather, dim=0)
+    return output     
+
+
+from typing import List
+def tie_encoder_decoder_weights(encoder: nn.Module, decoder: nn.Module, base_model_prefix: str, skip_key:str):
+    uninitialized_encoder_weights: List[str] = []
+    if decoder.__class__ != encoder.__class__:
+        print(
+            f"{decoder.__class__} and {encoder.__class__} are not equal. In this case make sure that all encoder weights are correctly initialized."
+        )
+
+    def tie_encoder_to_decoder_recursively(
+        decoder_pointer: nn.Module,
+        encoder_pointer: nn.Module,
+        module_name: str,
+        uninitialized_encoder_weights: List[str],
+        skip_key: str,
+        depth=0,
+    ):
+        assert isinstance(decoder_pointer, nn.Module) and isinstance(
+            encoder_pointer, nn.Module
+        ), f"{decoder_pointer} and {encoder_pointer} have to be of type torch.nn.Module"
+        if hasattr(decoder_pointer, "weight") and skip_key not in module_name:
+            assert hasattr(encoder_pointer, "weight")
+            encoder_pointer.weight = decoder_pointer.weight
+            if hasattr(decoder_pointer, "bias"):
+                assert hasattr(encoder_pointer, "bias")
+                encoder_pointer.bias = decoder_pointer.bias                
+            print(module_name+' is tied')    
+            return
+
+        encoder_modules = encoder_pointer._modules
+        decoder_modules = decoder_pointer._modules
+        if len(decoder_modules) > 0:
+            assert (
+                len(encoder_modules) > 0
+            ), f"Encoder module {encoder_pointer} does not match decoder module {decoder_pointer}"
+
+            all_encoder_weights = set([module_name + "/" + sub_name for sub_name in encoder_modules.keys()])
+            encoder_layer_pos = 0
+            for name, module in decoder_modules.items():
+                if name.isdigit():
+                    encoder_name = str(int(name) + encoder_layer_pos)
+                    decoder_name = name
+                    if not isinstance(decoder_modules[decoder_name], type(encoder_modules[encoder_name])) and len(
+                        encoder_modules
+                    ) != len(decoder_modules):
+                        # this can happen if the name corresponds to the position in a list module list of layers
+                        # in this case the decoder has added a cross-attention that the encoder does not have
+                        # thus skip this step and subtract one layer pos from encoder
+                        encoder_layer_pos -= 1
+                        continue
+                elif name not in encoder_modules:
+                    continue
+                elif depth > 500:
+                    raise ValueError(
+                        "Max depth of recursive function `tie_encoder_to_decoder` reached. It seems that there is a circular dependency between two or more `nn.Modules` of your model."
+                    )
+                else:
+                    decoder_name = encoder_name = name
+                tie_encoder_to_decoder_recursively(
+                    decoder_modules[decoder_name],
+                    encoder_modules[encoder_name],
+                    module_name + "/" + name,
+                    uninitialized_encoder_weights,
+                    skip_key,
+                    depth=depth + 1,
+                )
+                all_encoder_weights.remove(module_name + "/" + encoder_name)
+
+            uninitialized_encoder_weights += list(all_encoder_weights)
+
+    # tie weights recursively
+    tie_encoder_to_decoder_recursively(decoder, encoder, base_model_prefix, uninitialized_encoder_weights, skip_key)  

extras/BLIP/models/blip_retrieval.py

@@ -0,0 +1,319 @@
+from extras.BLIP.models.med import BertConfig, BertModel
+from transformers import BertTokenizer
+
+import torch
+from torch import nn
+import torch.nn.functional as F
+
+from extras.BLIP.models.blip import create_vit, init_tokenizer, load_checkpoint
+
+class BLIP_Retrieval(nn.Module):
+    def __init__(self,                 
+                 med_config = 'configs/med_config.json',  
+                 image_size = 384,
+                 vit = 'base',
+                 vit_grad_ckpt = False,
+                 vit_ckpt_layer = 0,                      
+                 embed_dim = 256,     
+                 queue_size = 57600,
+                 momentum = 0.995,
+                 negative_all_rank = False,
+                 ):
+        """
+        Args:
+            med_config (str): path for the mixture of encoder-decoder model's configuration file
+            image_size (int): input image size
+            vit (str): model size of vision transformer
+        """               
+        super().__init__()
+        
+        self.visual_encoder, vision_width = create_vit(vit,image_size, vit_grad_ckpt, vit_ckpt_layer)
+        self.tokenizer = init_tokenizer()   
+        med_config = BertConfig.from_json_file(med_config)
+        med_config.encoder_width = vision_width
+        self.text_encoder = BertModel(config=med_config, add_pooling_layer=False)          
+
+        text_width = self.text_encoder.config.hidden_size
+        
+        self.vision_proj = nn.Linear(vision_width, embed_dim)
+        self.text_proj = nn.Linear(text_width, embed_dim)
+
+        self.itm_head = nn.Linear(text_width, 2) 
+        
+        # create momentum encoders  
+        self.visual_encoder_m, vision_width = create_vit(vit,image_size)              
+        self.vision_proj_m = nn.Linear(vision_width, embed_dim)
+        self.text_encoder_m = BertModel(config=med_config, add_pooling_layer=False)    
+        self.text_proj_m = nn.Linear(text_width, embed_dim)
+        
+        self.model_pairs = [[self.visual_encoder,self.visual_encoder_m],
+                            [self.vision_proj,self.vision_proj_m],
+                            [self.text_encoder,self.text_encoder_m],
+                            [self.text_proj,self.text_proj_m],
+                           ]       
+        self.copy_params()
+
+        # create the queue
+        self.register_buffer("image_queue", torch.randn(embed_dim, queue_size))
+        self.register_buffer("text_queue", torch.randn(embed_dim, queue_size))
+        self.register_buffer("idx_queue", torch.full((1,queue_size),-100))
+        self.register_buffer("ptr_queue", torch.zeros(1, dtype=torch.long))  
+
+        self.image_queue = nn.functional.normalize(self.image_queue, dim=0)
+        self.text_queue = nn.functional.normalize(self.text_queue, dim=0)
+        
+        self.queue_size = queue_size
+        self.momentum = momentum
+        self.temp = nn.Parameter(0.07*torch.ones([]))   
+        
+        self.negative_all_rank = negative_all_rank
+        
+        
+    def forward(self, image, caption, alpha, idx):
+        with torch.no_grad():
+            self.temp.clamp_(0.001,0.5)
+        
+        image_embeds = self.visual_encoder(image) 
+        image_atts = torch.ones(image_embeds.size()[:-1],dtype=torch.long).to(image.device)        
+        image_feat = F.normalize(self.vision_proj(image_embeds[:,0,:]),dim=-1)    
+        
+        text = self.tokenizer(caption, padding='max_length', truncation=True, max_length=35, 
+                              return_tensors="pt").to(image.device) 
+        
+        text_output = self.text_encoder(text.input_ids, attention_mask = text.attention_mask,                      
+                                        return_dict = True, mode = 'text')            
+        text_feat = F.normalize(self.text_proj(text_output.last_hidden_state[:,0,:]),dim=-1)        
+        
+        ###============== Image-text Contrastive Learning ===================###
+        idx = idx.view(-1,1)
+        idx_all = torch.cat([idx.t(), self.idx_queue.clone().detach()],dim=1)  
+        pos_idx = torch.eq(idx, idx_all).float()       
+        sim_targets = pos_idx / pos_idx.sum(1,keepdim=True)   
+        
+        # get momentum features
+        with torch.no_grad():
+            self._momentum_update()
+            image_embeds_m = self.visual_encoder_m(image) 
+            image_feat_m = F.normalize(self.vision_proj_m(image_embeds_m[:,0,:]),dim=-1)  
+            image_feat_m_all = torch.cat([image_feat_m.t(),self.image_queue.clone().detach()],dim=1)                   
+            
+            text_output_m = self.text_encoder_m(text.input_ids, attention_mask = text.attention_mask,                      
+                                                return_dict = True, mode = 'text')    
+            text_feat_m = F.normalize(self.text_proj_m(text_output_m.last_hidden_state[:,0,:]),dim=-1) 
+            text_feat_m_all = torch.cat([text_feat_m.t(),self.text_queue.clone().detach()],dim=1)
+
+            sim_i2t_m = image_feat_m @ text_feat_m_all / self.temp  
+            sim_t2i_m = text_feat_m @ image_feat_m_all / self.temp   
+
+            sim_i2t_targets = alpha * F.softmax(sim_i2t_m, dim=1) + (1 - alpha) * sim_targets
+            sim_t2i_targets = alpha * F.softmax(sim_t2i_m, dim=1) + (1 - alpha) * sim_targets        
+
+        sim_i2t = image_feat @ text_feat_m_all / self.temp 
+        sim_t2i = text_feat @ image_feat_m_all / self.temp 
+                             
+        loss_i2t = -torch.sum(F.log_softmax(sim_i2t, dim=1)*sim_i2t_targets,dim=1).mean()
+        loss_t2i = -torch.sum(F.log_softmax(sim_t2i, dim=1)*sim_t2i_targets,dim=1).mean() 
+
+        loss_ita = (loss_i2t+loss_t2i)/2
+        
+        idxs = concat_all_gather(idx)
+        self._dequeue_and_enqueue(image_feat_m, text_feat_m, idxs)        
+
+        ###============== Image-text Matching ===================###
+        encoder_input_ids = text.input_ids.clone()
+        encoder_input_ids[:,0] = self.tokenizer.enc_token_id
+
+        # forward the positve image-text pair
+        bs = image.size(0)
+        output_pos = self.text_encoder(encoder_input_ids,
+                                       attention_mask = text.attention_mask,
+                                       encoder_hidden_states = image_embeds,
+                                       encoder_attention_mask = image_atts,      
+                                       return_dict = True,
+                                      )  
+        
+        
+        if self.negative_all_rank:    
+            # compute sample similarity
+            with torch.no_grad():                
+                mask = torch.eq(idx, idxs.t())
+
+                image_feat_world = concat_all_gather(image_feat)
+                text_feat_world = concat_all_gather(text_feat)
+
+                sim_i2t = image_feat @ text_feat_world.t() / self.temp 
+                sim_t2i = text_feat @ image_feat_world.t() / self.temp 
+
+                weights_i2t = F.softmax(sim_i2t,dim=1)
+                weights_i2t.masked_fill_(mask, 0)            
+
+                weights_t2i = F.softmax(sim_t2i,dim=1)
+                weights_t2i.masked_fill_(mask, 0)     
+
+            image_embeds_world = all_gather_with_grad(image_embeds) 
+
+            # select a negative image (from all ranks) for each text
+            image_embeds_neg = []    
+            for b in range(bs):
+                neg_idx = torch.multinomial(weights_t2i[b], 1).item()
+                image_embeds_neg.append(image_embeds_world[neg_idx])
+            image_embeds_neg = torch.stack(image_embeds_neg,dim=0)   
+
+            # select a negative text (from all ranks) for each image
+            input_ids_world = concat_all_gather(encoder_input_ids)
+            att_mask_world = concat_all_gather(text.attention_mask)        
+
+            text_ids_neg = []
+            text_atts_neg = []
+            for b in range(bs):
+                neg_idx = torch.multinomial(weights_i2t[b], 1).item()
+                text_ids_neg.append(input_ids_world[neg_idx])
+                text_atts_neg.append(att_mask_world[neg_idx])
+                
+        else:
+            with torch.no_grad():                
+                mask = torch.eq(idx, idx.t())
+                
+                sim_i2t = image_feat @ text_feat.t() / self.temp 
+                sim_t2i = text_feat @ image_feat.t() / self.temp 
+
+                weights_i2t = F.softmax(sim_i2t,dim=1)
+                weights_i2t.masked_fill_(mask, 0)            
+
+                weights_t2i = F.softmax(sim_t2i,dim=1)
+                weights_t2i.masked_fill_(mask, 0)     
+
+            # select a negative image (from same rank) for each text
+            image_embeds_neg = []    
+            for b in range(bs):
+                neg_idx = torch.multinomial(weights_t2i[b], 1).item()
+                image_embeds_neg.append(image_embeds[neg_idx])
+            image_embeds_neg = torch.stack(image_embeds_neg,dim=0)   
+
+            # select a negative text (from same rank) for each image    
+            text_ids_neg = []
+            text_atts_neg = []
+            for b in range(bs):
+                neg_idx = torch.multinomial(weights_i2t[b], 1).item()
+                text_ids_neg.append(encoder_input_ids[neg_idx])
+                text_atts_neg.append(text.attention_mask[neg_idx])            
+            
+        text_ids_neg = torch.stack(text_ids_neg,dim=0)   
+        text_atts_neg = torch.stack(text_atts_neg,dim=0)      
+
+        text_ids_all = torch.cat([encoder_input_ids, text_ids_neg],dim=0)     
+        text_atts_all = torch.cat([text.attention_mask, text_atts_neg],dim=0)     
+
+        image_embeds_all = torch.cat([image_embeds_neg,image_embeds],dim=0)
+        image_atts_all = torch.cat([image_atts,image_atts],dim=0)
+
+        output_neg = self.text_encoder(text_ids_all,
+                                       attention_mask = text_atts_all,
+                                       encoder_hidden_states = image_embeds_all,
+                                       encoder_attention_mask = image_atts_all,      
+                                       return_dict = True,
+                                      )                         
+          
+
+        vl_embeddings = torch.cat([output_pos.last_hidden_state[:,0,:], output_neg.last_hidden_state[:,0,:]],dim=0)
+        vl_output = self.itm_head(vl_embeddings)            
+
+        itm_labels = torch.cat([torch.ones(bs,dtype=torch.long),torch.zeros(2*bs,dtype=torch.long)],
+                               dim=0).to(image.device)
+        loss_itm = F.cross_entropy(vl_output, itm_labels)     
+
+        return loss_ita, loss_itm 
+ 
+
+    @torch.no_grad()    
+    def copy_params(self):
+        for model_pair in self.model_pairs:           
+            for param, param_m in zip(model_pair[0].parameters(), model_pair[1].parameters()):
+                param_m.data.copy_(param.data)  # initialize
+                param_m.requires_grad = False  # not update by gradient    
+
+            
+    @torch.no_grad()        
+    def _momentum_update(self):
+        for model_pair in self.model_pairs:           
+            for param, param_m in zip(model_pair[0].parameters(), model_pair[1].parameters()):
+                param_m.data = param_m.data * self.momentum + param.data * (1. - self.momentum)
+                
+                
+    @torch.no_grad()
+    def _dequeue_and_enqueue(self, image_feat, text_feat, idxs):
+        # gather keys before updating queue
+        image_feats = concat_all_gather(image_feat)
+        text_feats = concat_all_gather(text_feat)
+        
+
+        batch_size = image_feats.shape[0]
+
+        ptr = int(self.ptr_queue)
+        assert self.queue_size % batch_size == 0  # for simplicity
+
+        # replace the keys at ptr (dequeue and enqueue)
+        self.image_queue[:, ptr:ptr + batch_size] = image_feats.T
+        self.text_queue[:, ptr:ptr + batch_size] = text_feats.T
+        self.idx_queue[:, ptr:ptr + batch_size] = idxs.T
+        ptr = (ptr + batch_size) % self.queue_size # move pointer
+
+        self.ptr_queue[0] = ptr  
+
+
+def blip_retrieval(pretrained='',**kwargs):
+    model = BLIP_Retrieval(**kwargs)
+    if pretrained:
+        model,msg = load_checkpoint(model,pretrained)
+        print("missing keys:")
+        print(msg.missing_keys)
+    return model 
+
+
+@torch.no_grad()
+def concat_all_gather(tensor):
+    """
+    Performs all_gather operation on the provided tensors.
+    *** Warning ***: torch.distributed.all_gather has no gradient.
+    """
+    tensors_gather = [torch.ones_like(tensor)
+        for _ in range(torch.distributed.get_world_size())]
+    torch.distributed.all_gather(tensors_gather, tensor, async_op=False)
+
+    output = torch.cat(tensors_gather, dim=0)
+    return output      
+
+
+class GatherLayer(torch.autograd.Function):
+    """
+    Gather tensors from all workers with support for backward propagation:
+    This implementation does not cut the gradients as torch.distributed.all_gather does.
+    """
+
+    @staticmethod
+    def forward(ctx, x):
+        output = [torch.zeros_like(x) for _ in range(torch.distributed.get_world_size())]
+        torch.distributed.all_gather(output, x)
+        return tuple(output)
+
+    @staticmethod
+    def backward(ctx, *grads):
+        all_gradients = torch.stack(grads)
+        torch.distributed.all_reduce(all_gradients)
+        return all_gradients[torch.distributed.get_rank()]
+
+
+def all_gather_with_grad(tensors):
+    """
+    Performs all_gather operation on the provided tensors.
+    Graph remains connected for backward grad computation.
+    """
+    # Queue the gathered tensors
+    world_size = torch.distributed.get_world_size()
+    # There is no need for reduction in the single-proc case
+    if world_size == 1:
+        return tensors
+
+    tensor_all = GatherLayer.apply(tensors)
+
+    return torch.cat(tensor_all, dim=0)

extras/BLIP/models/blip_vqa.py

@@ -0,0 +1,186 @@
+from extras.BLIP.models.med import BertConfig, BertModel, BertLMHeadModel
+from extras.BLIP.models.blip import create_vit, init_tokenizer, load_checkpoint
+
+import torch
+from torch import nn
+import torch.nn.functional as F
+from transformers import BertTokenizer
+import numpy as np
+
+class BLIP_VQA(nn.Module):
+    def __init__(self,                 
+                 med_config = 'configs/med_config.json',  
+                 image_size = 480,
+                 vit = 'base',
+                 vit_grad_ckpt = False,
+                 vit_ckpt_layer = 0,                   
+                 ):
+        """
+        Args:
+            med_config (str): path for the mixture of encoder-decoder model's configuration file
+            image_size (int): input image size
+            vit (str): model size of vision transformer
+        """               
+        super().__init__()
+        
+        self.visual_encoder, vision_width = create_vit(vit, image_size, vit_grad_ckpt, vit_ckpt_layer, drop_path_rate=0.1)
+        self.tokenizer = init_tokenizer()  
+        
+        encoder_config = BertConfig.from_json_file(med_config)
+        encoder_config.encoder_width = vision_width
+        self.text_encoder = BertModel(config=encoder_config, add_pooling_layer=False) 
+        
+        decoder_config = BertConfig.from_json_file(med_config)        
+        self.text_decoder = BertLMHeadModel(config=decoder_config)          
+
+
+    def forward(self, image, question, answer=None, n=None, weights=None, train=True, inference='rank', k_test=128):
+        
+        image_embeds = self.visual_encoder(image) 
+        image_atts = torch.ones(image_embeds.size()[:-1],dtype=torch.long).to(image.device)
+        
+        question = self.tokenizer(question, padding='longest', truncation=True, max_length=35, 
+                                  return_tensors="pt").to(image.device) 
+        question.input_ids[:,0] = self.tokenizer.enc_token_id
+        
+        if train:               
+            '''
+            n: number of answers for each question
+            weights: weight for each answer
+            '''                     
+            answer = self.tokenizer(answer, padding='longest', return_tensors="pt").to(image.device) 
+            answer.input_ids[:,0] = self.tokenizer.bos_token_id
+            answer_targets = answer.input_ids.masked_fill(answer.input_ids == self.tokenizer.pad_token_id, -100)      
+
+            question_output = self.text_encoder(question.input_ids, 
+                                                attention_mask = question.attention_mask, 
+                                                encoder_hidden_states = image_embeds,
+                                                encoder_attention_mask = image_atts,                             
+                                                return_dict = True)    
+
+            question_states = []                
+            question_atts = []  
+            for b, n in enumerate(n):
+                question_states += [question_output.last_hidden_state[b]]*n
+                question_atts += [question.attention_mask[b]]*n                
+            question_states = torch.stack(question_states,0)    
+            question_atts = torch.stack(question_atts,0)     
+
+            answer_output = self.text_decoder(answer.input_ids, 
+                                              attention_mask = answer.attention_mask, 
+                                              encoder_hidden_states = question_states,
+                                              encoder_attention_mask = question_atts,                  
+                                              labels = answer_targets,
+                                              return_dict = True,   
+                                              reduction = 'none',
+                                             )      
+            
+            loss = weights * answer_output.loss
+            loss = loss.sum()/image.size(0)
+
+            return loss
+            
+
+        else: 
+            question_output = self.text_encoder(question.input_ids, 
+                                                attention_mask = question.attention_mask, 
+                                                encoder_hidden_states = image_embeds,
+                                                encoder_attention_mask = image_atts,                                    
+                                                return_dict = True) 
+            
+            if inference=='generate':
+                num_beams = 3
+                question_states = question_output.last_hidden_state.repeat_interleave(num_beams,dim=0)
+                question_atts = torch.ones(question_states.size()[:-1],dtype=torch.long).to(question_states.device)
+                model_kwargs = {"encoder_hidden_states": question_states, "encoder_attention_mask":question_atts}
+                
+                bos_ids = torch.full((image.size(0),1),fill_value=self.tokenizer.bos_token_id,device=image.device)
+                
+                outputs = self.text_decoder.generate(input_ids=bos_ids,
+                                                     max_length=10,
+                                                     min_length=1,
+                                                     num_beams=num_beams,
+                                                     eos_token_id=self.tokenizer.sep_token_id,
+                                                     pad_token_id=self.tokenizer.pad_token_id, 
+                                                     **model_kwargs)
+                
+                answers = []    
+                for output in outputs:
+                    answer = self.tokenizer.decode(output, skip_special_tokens=True)    
+                    answers.append(answer)
+                return answers
+            
+            elif inference=='rank':
+                max_ids = self.rank_answer(question_output.last_hidden_state, question.attention_mask, 
+                                           answer.input_ids, answer.attention_mask, k_test) 
+                return max_ids
+ 
+                
+                
+    def rank_answer(self, question_states, question_atts, answer_ids, answer_atts, k):
+        
+        num_ques = question_states.size(0)
+        start_ids = answer_ids[0,0].repeat(num_ques,1) # bos token
+        
+        start_output = self.text_decoder(start_ids, 
+                                         encoder_hidden_states = question_states,
+                                         encoder_attention_mask = question_atts,                                      
+                                         return_dict = True,
+                                         reduction = 'none')              
+        logits = start_output.logits[:,0,:] # first token's logit
+        
+        # topk_probs: top-k probability 
+        # topk_ids: [num_question, k]        
+        answer_first_token = answer_ids[:,1]
+        prob_first_token = F.softmax(logits,dim=1).index_select(dim=1, index=answer_first_token) 
+        topk_probs, topk_ids = prob_first_token.topk(k,dim=1) 
+        
+        # answer input: [num_question*k, answer_len]                 
+        input_ids = []
+        input_atts = []
+        for b, topk_id in enumerate(topk_ids):
+            input_ids.append(answer_ids.index_select(dim=0, index=topk_id))
+            input_atts.append(answer_atts.index_select(dim=0, index=topk_id))
+        input_ids = torch.cat(input_ids,dim=0)  
+        input_atts = torch.cat(input_atts,dim=0)  
+
+        targets_ids = input_ids.masked_fill(input_ids == self.tokenizer.pad_token_id, -100)
+
+        # repeat encoder's output for top-k answers
+        question_states = tile(question_states, 0, k)
+        question_atts = tile(question_atts, 0, k)
+        
+        output = self.text_decoder(input_ids, 
+                                   attention_mask = input_atts, 
+                                   encoder_hidden_states = question_states,
+                                   encoder_attention_mask = question_atts,     
+                                   labels = targets_ids,
+                                   return_dict = True, 
+                                   reduction = 'none')   
+        
+        log_probs_sum = -output.loss
+        log_probs_sum = log_probs_sum.view(num_ques,k)
+
+        max_topk_ids = log_probs_sum.argmax(dim=1) 
+        max_ids = topk_ids[max_topk_ids>=0,max_topk_ids]
+
+        return max_ids
+    
+    
+def blip_vqa(pretrained='',**kwargs):
+    model = BLIP_VQA(**kwargs)
+    if pretrained:
+        model,msg = load_checkpoint(model,pretrained)
+#         assert(len(msg.missing_keys)==0)
+    return model  
+
+
+def tile(x, dim, n_tile):
+    init_dim = x.size(dim)
+    repeat_idx = [1] * x.dim()
+    repeat_idx[dim] = n_tile
+    x = x.repeat(*(repeat_idx))
+    order_index = torch.LongTensor(np.concatenate([init_dim * np.arange(n_tile) + i for i in range(init_dim)]))
+    return torch.index_select(x, dim, order_index.to(x.device))    
+        
+        

extras/BLIP/models/med.py

@@ -0,0 +1,955 @@
+'''
+ * Copyright (c) 2022, salesforce.com, inc.
+ * All rights reserved.
+ * SPDX-License-Identifier: BSD-3-Clause
+ * For full license text, see LICENSE.txt file in the repo root or https://opensource.org/licenses/BSD-3-Clause
+ * By Junnan Li
+ * Based on huggingface code base
+ * https://github.com/huggingface/transformers/blob/v4.15.0/src/transformers/models/bert
+'''
+
+import math
+import os
+import warnings
+from dataclasses import dataclass
+from typing import Optional, Tuple
+
+import torch
+from torch import Tensor, device, dtype, nn
+import torch.utils.checkpoint
+from torch import nn
+from torch.nn import CrossEntropyLoss
+import torch.nn.functional as F
+
+from transformers.activations import ACT2FN
+from transformers.file_utils import (
+    ModelOutput,
+)
+from transformers.modeling_outputs import (
+    BaseModelOutputWithPastAndCrossAttentions,
+    BaseModelOutputWithPoolingAndCrossAttentions,
+    CausalLMOutputWithCrossAttentions,
+    MaskedLMOutput,
+    MultipleChoiceModelOutput,
+    NextSentencePredictorOutput,
+    QuestionAnsweringModelOutput,
+    SequenceClassifierOutput,
+    TokenClassifierOutput,
+)
+from transformers.modeling_utils import (
+    PreTrainedModel,
+    apply_chunking_to_forward,
+    find_pruneable_heads_and_indices,
+    prune_linear_layer,
+)
+from transformers.utils import logging
+from transformers.models.bert.configuration_bert import BertConfig
+
+
+logger = logging.get_logger(__name__)
+
+
+class BertEmbeddings(nn.Module):
+    """Construct the embeddings from word and position embeddings."""
+
+    def __init__(self, config):
+        super().__init__()
+        self.word_embeddings = nn.Embedding(config.vocab_size, config.hidden_size, padding_idx=config.pad_token_id)
+        self.position_embeddings = nn.Embedding(config.max_position_embeddings, config.hidden_size)
+
+        # self.LayerNorm is not snake-cased to stick with TensorFlow model variable name and be able to load
+        # any TensorFlow checkpoint file
+        self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
+        self.dropout = nn.Dropout(config.hidden_dropout_prob)
+
+        # position_ids (1, len position emb) is contiguous in memory and exported when serialized
+        self.register_buffer("position_ids", torch.arange(config.max_position_embeddings).expand((1, -1)))
+        self.position_embedding_type = getattr(config, "position_embedding_type", "absolute")
+        
+        self.config = config
+
+    def forward(
+        self, input_ids=None, position_ids=None, inputs_embeds=None, past_key_values_length=0
+    ):
+        if input_ids is not None:
+            input_shape = input_ids.size()
+        else:
+            input_shape = inputs_embeds.size()[:-1]
+
+        seq_length = input_shape[1]
+
+        if position_ids is None:
+            position_ids = self.position_ids[:, past_key_values_length : seq_length + past_key_values_length]
+
+        if inputs_embeds is None:
+            inputs_embeds = self.word_embeddings(input_ids)
+
+        embeddings = inputs_embeds
+
+        if self.position_embedding_type == "absolute":
+            position_embeddings = self.position_embeddings(position_ids)
+            embeddings += position_embeddings
+        embeddings = self.LayerNorm(embeddings)
+        embeddings = self.dropout(embeddings)
+        return embeddings
+
+
+class BertSelfAttention(nn.Module):
+    def __init__(self, config, is_cross_attention):
+        super().__init__()
+        self.config = config
+        if config.hidden_size % config.num_attention_heads != 0 and not hasattr(config, "embedding_size"):
+            raise ValueError(
+                "The hidden size (%d) is not a multiple of the number of attention "
+                "heads (%d)" % (config.hidden_size, config.num_attention_heads)
+            )
+        
+        self.num_attention_heads = config.num_attention_heads
+        self.attention_head_size = int(config.hidden_size / config.num_attention_heads)
+        self.all_head_size = self.num_attention_heads * self.attention_head_size
+
+        self.query = nn.Linear(config.hidden_size, self.all_head_size)
+        if is_cross_attention:
+            self.key = nn.Linear(config.encoder_width, self.all_head_size)
+            self.value = nn.Linear(config.encoder_width, self.all_head_size)
+        else:
+            self.key = nn.Linear(config.hidden_size, self.all_head_size)
+            self.value = nn.Linear(config.hidden_size, self.all_head_size)
+
+        self.dropout = nn.Dropout(config.attention_probs_dropout_prob)
+        self.position_embedding_type = getattr(config, "position_embedding_type", "absolute")
+        if self.position_embedding_type == "relative_key" or self.position_embedding_type == "relative_key_query":
+            self.max_position_embeddings = config.max_position_embeddings
+            self.distance_embedding = nn.Embedding(2 * config.max_position_embeddings - 1, self.attention_head_size)
+        self.save_attention = False   
+            
+    def save_attn_gradients(self, attn_gradients):
+        self.attn_gradients = attn_gradients
+        
+    def get_attn_gradients(self):
+        return self.attn_gradients
+    
+    def save_attention_map(self, attention_map):
+        self.attention_map = attention_map
+        
+    def get_attention_map(self):
+        return self.attention_map
+    
+    def transpose_for_scores(self, x):
+        new_x_shape = x.size()[:-1] + (self.num_attention_heads, self.attention_head_size)
+        x = x.view(*new_x_shape)
+        return x.permute(0, 2, 1, 3)
+
+    def forward(
+        self,
+        hidden_states,
+        attention_mask=None,
+        head_mask=None,
+        encoder_hidden_states=None,
+        encoder_attention_mask=None,
+        past_key_value=None,
+        output_attentions=False,
+    ):
+        mixed_query_layer = self.query(hidden_states)
+
+        # If this is instantiated as a cross-attention module, the keys
+        # and values come from an encoder; the attention mask needs to be
+        # such that the encoder's padding tokens are not attended to.
+        is_cross_attention = encoder_hidden_states is not None
+
+        if is_cross_attention:
+            key_layer = self.transpose_for_scores(self.key(encoder_hidden_states))
+            value_layer = self.transpose_for_scores(self.value(encoder_hidden_states))
+            attention_mask = encoder_attention_mask
+        elif past_key_value is not None:
+            key_layer = self.transpose_for_scores(self.key(hidden_states))
+            value_layer = self.transpose_for_scores(self.value(hidden_states))
+            key_layer = torch.cat([past_key_value[0], key_layer], dim=2)
+            value_layer = torch.cat([past_key_value[1], value_layer], dim=2)
+        else:
+            key_layer = self.transpose_for_scores(self.key(hidden_states))
+            value_layer = self.transpose_for_scores(self.value(hidden_states))
+
+        query_layer = self.transpose_for_scores(mixed_query_layer)
+
+        past_key_value = (key_layer, value_layer)
+
+        # Take the dot product between "query" and "key" to get the raw attention scores.
+        attention_scores = torch.matmul(query_layer, key_layer.transpose(-1, -2))
+
+        if self.position_embedding_type == "relative_key" or self.position_embedding_type == "relative_key_query":
+            seq_length = hidden_states.size()[1]
+            position_ids_l = torch.arange(seq_length, dtype=torch.long, device=hidden_states.device).view(-1, 1)
+            position_ids_r = torch.arange(seq_length, dtype=torch.long, device=hidden_states.device).view(1, -1)
+            distance = position_ids_l - position_ids_r
+            positional_embedding = self.distance_embedding(distance + self.max_position_embeddings - 1)
+            positional_embedding = positional_embedding.to(dtype=query_layer.dtype)  # fp16 compatibility
+
+            if self.position_embedding_type == "relative_key":
+                relative_position_scores = torch.einsum("bhld,lrd->bhlr", query_layer, positional_embedding)
+                attention_scores = attention_scores + relative_position_scores
+            elif self.position_embedding_type == "relative_key_query":
+                relative_position_scores_query = torch.einsum("bhld,lrd->bhlr", query_layer, positional_embedding)
+                relative_position_scores_key = torch.einsum("bhrd,lrd->bhlr", key_layer, positional_embedding)
+                attention_scores = attention_scores + relative_position_scores_query + relative_position_scores_key
+
+        attention_scores = attention_scores / math.sqrt(self.attention_head_size)
+        if attention_mask is not None:
+            # Apply the attention mask is (precomputed for all layers in BertModel forward() function)
+            attention_scores = attention_scores + attention_mask
+
+        # Normalize the attention scores to probabilities.
+        attention_probs = nn.Softmax(dim=-1)(attention_scores)
+        
+        if is_cross_attention and self.save_attention:
+            self.save_attention_map(attention_probs)
+            attention_probs.register_hook(self.save_attn_gradients)         
+
+        # This is actually dropping out entire tokens to attend to, which might
+        # seem a bit unusual, but is taken from the original Transformer paper.
+        attention_probs_dropped = self.dropout(attention_probs)
+
+        # Mask heads if we want to
+        if head_mask is not None:
+            attention_probs_dropped = attention_probs_dropped * head_mask
+
+        context_layer = torch.matmul(attention_probs_dropped, value_layer)
+
+        context_layer = context_layer.permute(0, 2, 1, 3).contiguous()
+        new_context_layer_shape = context_layer.size()[:-2] + (self.all_head_size,)
+        context_layer = context_layer.view(*new_context_layer_shape)
+
+        outputs = (context_layer, attention_probs) if output_attentions else (context_layer,)
+
+        outputs = outputs + (past_key_value,)
+        return outputs
+
+
+class BertSelfOutput(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.dense = nn.Linear(config.hidden_size, config.hidden_size)
+        self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
+        self.dropout = nn.Dropout(config.hidden_dropout_prob)
+
+    def forward(self, hidden_states, input_tensor):
+        hidden_states = self.dense(hidden_states)
+        hidden_states = self.dropout(hidden_states)
+        hidden_states = self.LayerNorm(hidden_states + input_tensor)
+        return hidden_states
+
+
+class BertAttention(nn.Module):
+    def __init__(self, config, is_cross_attention=False):
+        super().__init__()
+        self.self = BertSelfAttention(config, is_cross_attention)
+        self.output = BertSelfOutput(config)
+        self.pruned_heads = set()
+
+    def prune_heads(self, heads):
+        if len(heads) == 0:
+            return
+        heads, index = find_pruneable_heads_and_indices(
+            heads, self.self.num_attention_heads, self.self.attention_head_size, self.pruned_heads
+        )
+
+        # Prune linear layers
+        self.self.query = prune_linear_layer(self.self.query, index)
+        self.self.key = prune_linear_layer(self.self.key, index)
+        self.self.value = prune_linear_layer(self.self.value, index)
+        self.output.dense = prune_linear_layer(self.output.dense, index, dim=1)
+
+        # Update hyper params and store pruned heads
+        self.self.num_attention_heads = self.self.num_attention_heads - len(heads)
+        self.self.all_head_size = self.self.attention_head_size * self.self.num_attention_heads
+        self.pruned_heads = self.pruned_heads.union(heads)
+
+    def forward(
+        self,
+        hidden_states,
+        attention_mask=None,
+        head_mask=None,
+        encoder_hidden_states=None,
+        encoder_attention_mask=None,
+        past_key_value=None,
+        output_attentions=False,
+    ):
+        self_outputs = self.self(
+            hidden_states,
+            attention_mask,
+            head_mask,
+            encoder_hidden_states,
+            encoder_attention_mask,
+            past_key_value,
+            output_attentions,
+        )
+        attention_output = self.output(self_outputs[0], hidden_states)
+        outputs = (attention_output,) + self_outputs[1:]  # add attentions if we output them
+        return outputs
+
+
+class BertIntermediate(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.dense = nn.Linear(config.hidden_size, config.intermediate_size)
+        if isinstance(config.hidden_act, str):
+            self.intermediate_act_fn = ACT2FN[config.hidden_act]
+        else:
+            self.intermediate_act_fn = config.hidden_act
+
+    def forward(self, hidden_states):
+        hidden_states = self.dense(hidden_states)
+        hidden_states = self.intermediate_act_fn(hidden_states)
+        return hidden_states
+
+
+class BertOutput(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.dense = nn.Linear(config.intermediate_size, config.hidden_size)
+        self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
+        self.dropout = nn.Dropout(config.hidden_dropout_prob)
+
+    def forward(self, hidden_states, input_tensor):
+        hidden_states = self.dense(hidden_states)
+        hidden_states = self.dropout(hidden_states)
+        hidden_states = self.LayerNorm(hidden_states + input_tensor)
+        return hidden_states
+
+
+class BertLayer(nn.Module):
+    def __init__(self, config, layer_num):
+        super().__init__()
+        self.config = config
+        self.chunk_size_feed_forward = config.chunk_size_feed_forward
+        self.seq_len_dim = 1
+        self.attention = BertAttention(config)      
+        self.layer_num = layer_num          
+        if self.config.add_cross_attention:
+            self.crossattention = BertAttention(config, is_cross_attention=self.config.add_cross_attention)
+        self.intermediate = BertIntermediate(config)
+        self.output = BertOutput(config)
+
+    def forward(
+        self,
+        hidden_states,
+        attention_mask=None,
+        head_mask=None,
+        encoder_hidden_states=None,
+        encoder_attention_mask=None,
+        past_key_value=None,
+        output_attentions=False,
+        mode=None,
+    ):
+        # decoder uni-directional self-attention cached key/values tuple is at positions 1,2
+        self_attn_past_key_value = past_key_value[:2] if past_key_value is not None else None
+        self_attention_outputs = self.attention(
+            hidden_states,
+            attention_mask,
+            head_mask,
+            output_attentions=output_attentions,
+            past_key_value=self_attn_past_key_value,
+        )
+        attention_output = self_attention_outputs[0]
+
+        outputs = self_attention_outputs[1:-1]
+        present_key_value = self_attention_outputs[-1]
+
+        if mode=='multimodal':
+            assert encoder_hidden_states is not None, "encoder_hidden_states must be given for cross-attention layers"
+
+            cross_attention_outputs = self.crossattention(
+                attention_output,
+                attention_mask,
+                head_mask,
+                encoder_hidden_states,
+                encoder_attention_mask,
+                output_attentions=output_attentions,
+            )
+            attention_output = cross_attention_outputs[0]
+            outputs = outputs + cross_attention_outputs[1:-1]  # add cross attentions if we output attention weights                               
+        layer_output = apply_chunking_to_forward(
+            self.feed_forward_chunk, self.chunk_size_feed_forward, self.seq_len_dim, attention_output
+        )
+        outputs = (layer_output,) + outputs
+
+        outputs = outputs + (present_key_value,)
+
+        return outputs
+
+    def feed_forward_chunk(self, attention_output):
+        intermediate_output = self.intermediate(attention_output)
+        layer_output = self.output(intermediate_output, attention_output)
+        return layer_output
+
+
+class BertEncoder(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.config = config
+        self.layer = nn.ModuleList([BertLayer(config,i) for i in range(config.num_hidden_layers)])
+        self.gradient_checkpointing = False
+
+    def forward(
+        self,
+        hidden_states,
+        attention_mask=None,
+        head_mask=None,
+        encoder_hidden_states=None,
+        encoder_attention_mask=None,
+        past_key_values=None,
+        use_cache=None,
+        output_attentions=False,
+        output_hidden_states=False,
+        return_dict=True,
+        mode='multimodal',
+    ):
+        all_hidden_states = () if output_hidden_states else None
+        all_self_attentions = () if output_attentions else None
+        all_cross_attentions = () if output_attentions and self.config.add_cross_attention else None
+
+        next_decoder_cache = () if use_cache else None
+               
+        for i in range(self.config.num_hidden_layers):
+            layer_module = self.layer[i]
+            if output_hidden_states:
+                all_hidden_states = all_hidden_states + (hidden_states,)
+
+            layer_head_mask = head_mask[i] if head_mask is not None else None
+            past_key_value = past_key_values[i] if past_key_values is not None else None
+
+            if self.gradient_checkpointing and self.training:
+
+                if use_cache:
+                    logger.warn(
+                        "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
+                    )
+                    use_cache = False
+
+                def create_custom_forward(module):
+                    def custom_forward(*inputs):
+                        return module(*inputs, past_key_value, output_attentions)
+
+                    return custom_forward
+
+                layer_outputs = torch.utils.checkpoint.checkpoint(
+                    create_custom_forward(layer_module),
+                    hidden_states,
+                    attention_mask,
+                    layer_head_mask,
+                    encoder_hidden_states,
+                    encoder_attention_mask,
+                    mode=mode,
+                )
+            else:
+                layer_outputs = layer_module(
+                    hidden_states,
+                    attention_mask,
+                    layer_head_mask,
+                    encoder_hidden_states,
+                    encoder_attention_mask,
+                    past_key_value,
+                    output_attentions,
+                    mode=mode,
+                )
+
+            hidden_states = layer_outputs[0]
+            if use_cache:
+                next_decoder_cache += (layer_outputs[-1],)
+            if output_attentions:
+                all_self_attentions = all_self_attentions + (layer_outputs[1],)
+
+        if output_hidden_states:
+            all_hidden_states = all_hidden_states + (hidden_states,)
+
+        if not return_dict:
+            return tuple(
+                v
+                for v in [
+                    hidden_states,
+                    next_decoder_cache,
+                    all_hidden_states,
+                    all_self_attentions,
+                    all_cross_attentions,
+                ]
+                if v is not None
+            )
+        return BaseModelOutputWithPastAndCrossAttentions(
+            last_hidden_state=hidden_states,
+            past_key_values=next_decoder_cache,
+            hidden_states=all_hidden_states,
+            attentions=all_self_attentions,
+            cross_attentions=all_cross_attentions,
+        )
+
+
+class BertPooler(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.dense = nn.Linear(config.hidden_size, config.hidden_size)
+        self.activation = nn.Tanh()
+
+    def forward(self, hidden_states):
+        # We "pool" the model by simply taking the hidden state corresponding
+        # to the first token.
+        first_token_tensor = hidden_states[:, 0]
+        pooled_output = self.dense(first_token_tensor)
+        pooled_output = self.activation(pooled_output)
+        return pooled_output
+
+
+class BertPredictionHeadTransform(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.dense = nn.Linear(config.hidden_size, config.hidden_size)
+        if isinstance(config.hidden_act, str):
+            self.transform_act_fn = ACT2FN[config.hidden_act]
+        else:
+            self.transform_act_fn = config.hidden_act
+        self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
+
+    def forward(self, hidden_states):
+        hidden_states = self.dense(hidden_states)
+        hidden_states = self.transform_act_fn(hidden_states)
+        hidden_states = self.LayerNorm(hidden_states)
+        return hidden_states
+
+
+class BertLMPredictionHead(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.transform = BertPredictionHeadTransform(config)
+
+        # The output weights are the same as the input embeddings, but there is
+        # an output-only bias for each token.
+        self.decoder = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
+
+        self.bias = nn.Parameter(torch.zeros(config.vocab_size))
+
+        # Need a link between the two variables so that the bias is correctly resized with `resize_token_embeddings`
+        self.decoder.bias = self.bias
+
+    def forward(self, hidden_states):
+        hidden_states = self.transform(hidden_states)
+        hidden_states = self.decoder(hidden_states)
+        return hidden_states
+
+
+class BertOnlyMLMHead(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.predictions = BertLMPredictionHead(config)
+
+    def forward(self, sequence_output):
+        prediction_scores = self.predictions(sequence_output)
+        return prediction_scores
+
+
+class BertPreTrainedModel(PreTrainedModel):
+    """
+    An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained
+    models.
+    """
+
+    config_class = BertConfig
+    base_model_prefix = "bert"
+    _keys_to_ignore_on_load_missing = [r"position_ids"]
+
+    def _init_weights(self, module):
+        """ Initialize the weights """
+        if isinstance(module, (nn.Linear, nn.Embedding)):
+            # Slightly different from the TF version which uses truncated_normal for initialization
+            # cf https://github.com/pytorch/pytorch/pull/5617
+            module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)
+        elif isinstance(module, nn.LayerNorm):
+            module.bias.data.zero_()
+            module.weight.data.fill_(1.0)
+        if isinstance(module, nn.Linear) and module.bias is not None:
+            module.bias.data.zero_()
+
+
+class BertModel(BertPreTrainedModel):
+    """
+    The model can behave as an encoder (with only self-attention) as well as a decoder, in which case a layer of
+    cross-attention is added between the self-attention layers, following the architecture described in `Attention is
+    all you need <https://arxiv.org/abs/1706.03762>`__ by Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit,
+    Llion Jones, Aidan N. Gomez, Lukasz Kaiser and Illia Polosukhin.
+    argument and :obj:`add_cross_attention` set to :obj:`True`; an :obj:`encoder_hidden_states` is then expected as an
+    input to the forward pass.
+    """
+
+    def __init__(self, config, add_pooling_layer=True):
+        super().__init__(config)
+        self.config = config
+
+        self.embeddings = BertEmbeddings(config)
+        
+        self.encoder = BertEncoder(config)
+
+        self.pooler = BertPooler(config) if add_pooling_layer else None
+
+        self.init_weights()
+ 
+
+    def get_input_embeddings(self):
+        return self.embeddings.word_embeddings
+
+    def set_input_embeddings(self, value):
+        self.embeddings.word_embeddings = value
+
+    def _prune_heads(self, heads_to_prune):
+        """
+        Prunes heads of the model. heads_to_prune: dict of {layer_num: list of heads to prune in this layer} See base
+        class PreTrainedModel
+        """
+        for layer, heads in heads_to_prune.items():
+            self.encoder.layer[layer].attention.prune_heads(heads)
+
+    
+    def get_extended_attention_mask(self, attention_mask: Tensor, input_shape: Tuple[int], device: device, is_decoder: bool) -> Tensor:
+        """
+        Makes broadcastable attention and causal masks so that future and masked tokens are ignored.
+
+        Arguments:
+            attention_mask (:obj:`torch.Tensor`):
+                Mask with ones indicating tokens to attend to, zeros for tokens to ignore.
+            input_shape (:obj:`Tuple[int]`):
+                The shape of the input to the model.
+            device: (:obj:`torch.device`):
+                The device of the input to the model.
+
+        Returns:
+            :obj:`torch.Tensor` The extended attention mask, with a the same dtype as :obj:`attention_mask.dtype`.
+        """
+        # We can provide a self-attention mask of dimensions [batch_size, from_seq_length, to_seq_length]
+        # ourselves in which case we just need to make it broadcastable to all heads.
+        if attention_mask.dim() == 3:
+            extended_attention_mask = attention_mask[:, None, :, :]
+        elif attention_mask.dim() == 2:
+            # Provided a padding mask of dimensions [batch_size, seq_length]
+            # - if the model is a decoder, apply a causal mask in addition to the padding mask
+            # - if the model is an encoder, make the mask broadcastable to [batch_size, num_heads, seq_length, seq_length]
+            if is_decoder:
+                batch_size, seq_length = input_shape
+
+                seq_ids = torch.arange(seq_length, device=device)
+                causal_mask = seq_ids[None, None, :].repeat(batch_size, seq_length, 1) <= seq_ids[None, :, None]
+                # in case past_key_values are used we need to add a prefix ones mask to the causal mask
+                # causal and attention masks must have same type with pytorch version < 1.3
+                causal_mask = causal_mask.to(attention_mask.dtype)
+   
+                if causal_mask.shape[1] < attention_mask.shape[1]:
+                    prefix_seq_len = attention_mask.shape[1] - causal_mask.shape[1]
+                    causal_mask = torch.cat(
+                        [
+                            torch.ones((batch_size, seq_length, prefix_seq_len), device=device, dtype=causal_mask.dtype),
+                            causal_mask,
+                        ],
+                        axis=-1,
+                    )                     
+
+                extended_attention_mask = causal_mask[:, None, :, :] * attention_mask[:, None, None, :]
+            else:
+                extended_attention_mask = attention_mask[:, None, None, :]
+        else:
+            raise ValueError(
+                "Wrong shape for input_ids (shape {}) or attention_mask (shape {})".format(
+                    input_shape, attention_mask.shape
+                )
+            )
+
+        # Since attention_mask is 1.0 for positions we want to attend and 0.0 for
+        # masked positions, this operation will create a tensor which is 0.0 for
+        # positions we want to attend and -10000.0 for masked positions.
+        # Since we are adding it to the raw scores before the softmax, this is
+        # effectively the same as removing these entirely.
+        extended_attention_mask = extended_attention_mask.to(dtype=self.dtype)  # fp16 compatibility
+        extended_attention_mask = (1.0 - extended_attention_mask) * -10000.0
+        return extended_attention_mask
+    
+    def forward(
+        self,
+        input_ids=None,
+        attention_mask=None,
+        position_ids=None,
+        head_mask=None,
+        inputs_embeds=None,
+        encoder_embeds=None,
+        encoder_hidden_states=None,
+        encoder_attention_mask=None,
+        past_key_values=None,
+        use_cache=None,
+        output_attentions=None,
+        output_hidden_states=None,
+        return_dict=None,
+        is_decoder=False,
+        mode='multimodal',
+    ):
+        r"""
+        encoder_hidden_states  (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length, hidden_size)`, `optional`):
+            Sequence of hidden-states at the output of the last layer of the encoder. Used in the cross-attention if
+            the model is configured as a decoder.
+        encoder_attention_mask (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`):
+            Mask to avoid performing attention on the padding token indices of the encoder input. This mask is used in
+            the cross-attention if the model is configured as a decoder. Mask values selected in ``[0, 1]``:
+            - 1 for tokens that are **not masked**,
+            - 0 for tokens that are **masked**.
+        past_key_values (:obj:`tuple(tuple(torch.FloatTensor))` of length :obj:`config.n_layers` with each tuple having 4 tensors of shape :obj:`(batch_size, num_heads, sequence_length - 1, embed_size_per_head)`):
+            Contains precomputed key and value hidden states of the attention blocks. Can be used to speed up decoding.
+            If :obj:`past_key_values` are used, the user can optionally input only the last :obj:`decoder_input_ids`
+            (those that don't have their past key value states given to this model) of shape :obj:`(batch_size, 1)`
+            instead of all :obj:`decoder_input_ids` of shape :obj:`(batch_size, sequence_length)`.
+        use_cache (:obj:`bool`, `optional`):
+            If set to :obj:`True`, :obj:`past_key_values` key value states are returned and can be used to speed up
+            decoding (see :obj:`past_key_values`).
+        """
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        output_hidden_states = (
+            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        )
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        if is_decoder:
+            use_cache = use_cache if use_cache is not None else self.config.use_cache
+        else:
+            use_cache = False
+
+        if input_ids is not None and inputs_embeds is not None:
+            raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
+        elif input_ids is not None:
+            input_shape = input_ids.size()
+            batch_size, seq_length = input_shape
+            device = input_ids.device
+        elif inputs_embeds is not None:
+            input_shape = inputs_embeds.size()[:-1]
+            batch_size, seq_length = input_shape
+            device = inputs_embeds.device
+        elif encoder_embeds is not None:    
+            input_shape = encoder_embeds.size()[:-1]
+            batch_size, seq_length = input_shape 
+            device = encoder_embeds.device
+        else:
+            raise ValueError("You have to specify either input_ids or inputs_embeds or encoder_embeds")
+
+        # past_key_values_length
+        past_key_values_length = past_key_values[0][0].shape[2] if past_key_values is not None else 0
+
+        if attention_mask is None:
+            attention_mask = torch.ones(((batch_size, seq_length + past_key_values_length)), device=device)
+            
+        # We can provide a self-attention mask of dimensions [batch_size, from_seq_length, to_seq_length]
+        # ourselves in which case we just need to make it broadcastable to all heads.
+        extended_attention_mask: torch.Tensor = self.get_extended_attention_mask(attention_mask, input_shape, 
+                                                                                 device, is_decoder)
+
+        # If a 2D or 3D attention mask is provided for the cross-attention
+        # we need to make broadcastable to [batch_size, num_heads, seq_length, seq_length]
+        if encoder_hidden_states is not None:
+            if type(encoder_hidden_states) == list:
+                encoder_batch_size, encoder_sequence_length, _ = encoder_hidden_states[0].size()
+            else:
+                encoder_batch_size, encoder_sequence_length, _ = encoder_hidden_states.size()
+            encoder_hidden_shape = (encoder_batch_size, encoder_sequence_length)
+            
+            if type(encoder_attention_mask) == list:
+                encoder_extended_attention_mask = [self.invert_attention_mask(mask) for mask in encoder_attention_mask]
+            elif encoder_attention_mask is None:
+                encoder_attention_mask = torch.ones(encoder_hidden_shape, device=device)
+                encoder_extended_attention_mask = self.invert_attention_mask(encoder_attention_mask)
+            else:    
+                encoder_extended_attention_mask = self.invert_attention_mask(encoder_attention_mask)
+        else:
+            encoder_extended_attention_mask = None
+
+        # Prepare head mask if needed
+        # 1.0 in head_mask indicate we keep the head
+        # attention_probs has shape bsz x n_heads x N x N
+        # input head_mask has shape [num_heads] or [num_hidden_layers x num_heads]
+        # and head_mask is converted to shape [num_hidden_layers x batch x num_heads x seq_length x seq_length]
+        head_mask = self.get_head_mask(head_mask, self.config.num_hidden_layers)
+        
+        if encoder_embeds is None:
+            embedding_output = self.embeddings(
+                input_ids=input_ids,
+                position_ids=position_ids,
+                inputs_embeds=inputs_embeds,
+                past_key_values_length=past_key_values_length,
+            )
+        else:
+            embedding_output = encoder_embeds
+            
+        encoder_outputs = self.encoder(
+            embedding_output,
+            attention_mask=extended_attention_mask,
+            head_mask=head_mask,
+            encoder_hidden_states=encoder_hidden_states,
+            encoder_attention_mask=encoder_extended_attention_mask,
+            past_key_values=past_key_values,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+            mode=mode,
+        )
+        sequence_output = encoder_outputs[0]
+        pooled_output = self.pooler(sequence_output) if self.pooler is not None else None
+
+        if not return_dict:
+            return (sequence_output, pooled_output) + encoder_outputs[1:]
+
+        return BaseModelOutputWithPoolingAndCrossAttentions(
+            last_hidden_state=sequence_output,
+            pooler_output=pooled_output,
+            past_key_values=encoder_outputs.past_key_values,
+            hidden_states=encoder_outputs.hidden_states,
+            attentions=encoder_outputs.attentions,
+            cross_attentions=encoder_outputs.cross_attentions,
+        )
+
+
+
+class BertLMHeadModel(BertPreTrainedModel):
+
+    _keys_to_ignore_on_load_unexpected = [r"pooler"]
+    _keys_to_ignore_on_load_missing = [r"position_ids", r"predictions.decoder.bias"]
+
+    def __init__(self, config):
+        super().__init__(config)
+
+        self.bert = BertModel(config, add_pooling_layer=False)
+        self.cls = BertOnlyMLMHead(config)
+
+        self.init_weights()
+
+    def get_output_embeddings(self):
+        return self.cls.predictions.decoder
+
+    def set_output_embeddings(self, new_embeddings):
+        self.cls.predictions.decoder = new_embeddings
+
+    def forward(
+        self,
+        input_ids=None,
+        attention_mask=None,
+        position_ids=None,
+        head_mask=None,
+        inputs_embeds=None,
+        encoder_hidden_states=None,
+        encoder_attention_mask=None,
+        labels=None,
+        past_key_values=None,
+        use_cache=None,
+        output_attentions=None,
+        output_hidden_states=None,
+        return_dict=None,
+        return_logits=False,            
+        is_decoder=True,
+        reduction='mean',
+        mode='multimodal', 
+    ):
+        r"""
+        encoder_hidden_states  (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length, hidden_size)`, `optional`):
+            Sequence of hidden-states at the output of the last layer of the encoder. Used in the cross-attention if
+            the model is configured as a decoder.
+        encoder_attention_mask (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`):
+            Mask to avoid performing attention on the padding token indices of the encoder input. This mask is used in
+            the cross-attention if the model is configured as a decoder. Mask values selected in ``[0, 1]``:
+            - 1 for tokens that are **not masked**,
+            - 0 for tokens that are **masked**.
+        labels (:obj:`torch.LongTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`):
+            Labels for computing the left-to-right language modeling loss (next word prediction). Indices should be in
+            ``[-100, 0, ..., config.vocab_size]`` (see ``input_ids`` docstring) Tokens with indices set to ``-100`` are
+            ignored (masked), the loss is only computed for the tokens with labels n ``[0, ..., config.vocab_size]``
+        past_key_values (:obj:`tuple(tuple(torch.FloatTensor))` of length :obj:`config.n_layers` with each tuple having 4 tensors of shape :obj:`(batch_size, num_heads, sequence_length - 1, embed_size_per_head)`):
+            Contains precomputed key and value hidden states of the attention blocks. Can be used to speed up decoding.
+            If :obj:`past_key_values` are used, the user can optionally input only the last :obj:`decoder_input_ids`
+            (those that don't have their past key value states given to this model) of shape :obj:`(batch_size, 1)`
+            instead of all :obj:`decoder_input_ids` of shape :obj:`(batch_size, sequence_length)`.
+        use_cache (:obj:`bool`, `optional`):
+            If set to :obj:`True`, :obj:`past_key_values` key value states are returned and can be used to speed up
+            decoding (see :obj:`past_key_values`).
+        Returns:
+        Example::
+            >>> from transformers import BertTokenizer, BertLMHeadModel, BertConfig
+            >>> import torch
+            >>> tokenizer = BertTokenizer.from_pretrained('bert-base-cased')
+            >>> config = BertConfig.from_pretrained("bert-base-cased")
+            >>> model = BertLMHeadModel.from_pretrained('bert-base-cased', config=config)
+            >>> inputs = tokenizer("Hello, my dog is cute", return_tensors="pt")
+            >>> outputs = model(**inputs)
+            >>> prediction_logits = outputs.logits
+        """
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+        if labels is not None:
+            use_cache = False
+
+        outputs = self.bert(
+            input_ids,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            head_mask=head_mask,
+            inputs_embeds=inputs_embeds,
+            encoder_hidden_states=encoder_hidden_states,
+            encoder_attention_mask=encoder_attention_mask,
+            past_key_values=past_key_values,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+            is_decoder=is_decoder,
+            mode=mode,
+        )
+        
+        sequence_output = outputs[0]
+        prediction_scores = self.cls(sequence_output)
+        
+        if return_logits:
+            return prediction_scores[:, :-1, :].contiguous()  
+
+        lm_loss = None
+        if labels is not None:
+            # we are doing next-token prediction; shift prediction scores and input ids by one
+            shifted_prediction_scores = prediction_scores[:, :-1, :].contiguous()
+            labels = labels[:, 1:].contiguous()
+            loss_fct = CrossEntropyLoss(reduction=reduction, label_smoothing=0.1) 
+            lm_loss = loss_fct(shifted_prediction_scores.view(-1, self.config.vocab_size), labels.view(-1))
+            if reduction=='none':
+                lm_loss = lm_loss.view(prediction_scores.size(0),-1).sum(1)               
+
+        if not return_dict:
+            output = (prediction_scores,) + outputs[2:]
+            return ((lm_loss,) + output) if lm_loss is not None else output
+
+        return CausalLMOutputWithCrossAttentions(
+            loss=lm_loss,
+            logits=prediction_scores,
+            past_key_values=outputs.past_key_values,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+            cross_attentions=outputs.cross_attentions,
+        )
+
+    def prepare_inputs_for_generation(self, input_ids, past=None, attention_mask=None, **model_kwargs):
+        input_shape = input_ids.shape
+        # if model is used as a decoder in encoder-decoder model, the decoder attention mask is created on the fly
+        if attention_mask is None:
+            attention_mask = input_ids.new_ones(input_shape)
+
+        # cut decoder_input_ids if past is used
+        if past is not None:
+            input_ids = input_ids[:, -1:]
+
+        return {
+            "input_ids": input_ids, 
+            "attention_mask": attention_mask, 
+            "past_key_values": past,
+            "encoder_hidden_states": model_kwargs.get("encoder_hidden_states", None),
+            "encoder_attention_mask": model_kwargs.get("encoder_attention_mask", None),
+            "is_decoder": True,
+        }
+
+    def _reorder_cache(self, past, beam_idx):
+        reordered_past = ()
+        for layer_past in past:
+            reordered_past += (tuple(past_state.index_select(0, beam_idx) for past_state in layer_past),)
+        return reordered_past

extras/BLIP/models/nlvr_encoder.py

@@ -0,0 +1,843 @@
+import math
+import os
+import warnings
+from dataclasses import dataclass
+from typing import Optional, Tuple
+
+import torch
+from torch import Tensor, device, dtype, nn
+import torch.utils.checkpoint
+from torch import nn
+from torch.nn import CrossEntropyLoss
+import torch.nn.functional as F
+
+from transformers.activations import ACT2FN
+from transformers.file_utils import (
+    ModelOutput,
+)
+from transformers.modeling_outputs import (
+    BaseModelOutputWithPastAndCrossAttentions,
+    BaseModelOutputWithPoolingAndCrossAttentions,
+    CausalLMOutputWithCrossAttentions,
+    MaskedLMOutput,
+    MultipleChoiceModelOutput,
+    NextSentencePredictorOutput,
+    QuestionAnsweringModelOutput,
+    SequenceClassifierOutput,
+    TokenClassifierOutput,
+)
+from transformers.modeling_utils import (
+    PreTrainedModel,
+    apply_chunking_to_forward,
+    find_pruneable_heads_and_indices,
+    prune_linear_layer,
+)
+from transformers.utils import logging
+from transformers.models.bert.configuration_bert import BertConfig
+
+
+logger = logging.get_logger(__name__)
+
+
+class BertEmbeddings(nn.Module):
+    """Construct the embeddings from word and position embeddings."""
+
+    def __init__(self, config):
+        super().__init__()
+        self.word_embeddings = nn.Embedding(config.vocab_size, config.hidden_size, padding_idx=config.pad_token_id)
+        self.position_embeddings = nn.Embedding(config.max_position_embeddings, config.hidden_size)
+
+        # self.LayerNorm is not snake-cased to stick with TensorFlow model variable name and be able to load
+        # any TensorFlow checkpoint file
+        self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
+        self.dropout = nn.Dropout(config.hidden_dropout_prob)
+
+        # position_ids (1, len position emb) is contiguous in memory and exported when serialized
+        self.register_buffer("position_ids", torch.arange(config.max_position_embeddings).expand((1, -1)))
+        self.position_embedding_type = getattr(config, "position_embedding_type", "absolute")
+        
+        self.config = config
+
+    def forward(
+        self, input_ids=None, position_ids=None, inputs_embeds=None, past_key_values_length=0
+    ):
+        if input_ids is not None:
+            input_shape = input_ids.size()
+        else:
+            input_shape = inputs_embeds.size()[:-1]
+
+        seq_length = input_shape[1]
+
+        if position_ids is None:
+            position_ids = self.position_ids[:, past_key_values_length : seq_length + past_key_values_length]
+
+        if inputs_embeds is None:
+            inputs_embeds = self.word_embeddings(input_ids)
+
+        embeddings = inputs_embeds
+
+        if self.position_embedding_type == "absolute":
+            position_embeddings = self.position_embeddings(position_ids)
+            embeddings += position_embeddings
+        embeddings = self.LayerNorm(embeddings)
+        embeddings = self.dropout(embeddings)
+        return embeddings
+
+
+class BertSelfAttention(nn.Module):
+    def __init__(self, config, is_cross_attention):
+        super().__init__()
+        self.config = config
+        if config.hidden_size % config.num_attention_heads != 0 and not hasattr(config, "embedding_size"):
+            raise ValueError(
+                "The hidden size (%d) is not a multiple of the number of attention "
+                "heads (%d)" % (config.hidden_size, config.num_attention_heads)
+            )
+        
+        self.num_attention_heads = config.num_attention_heads
+        self.attention_head_size = int(config.hidden_size / config.num_attention_heads)
+        self.all_head_size = self.num_attention_heads * self.attention_head_size
+
+        self.query = nn.Linear(config.hidden_size, self.all_head_size)
+        if is_cross_attention:
+            self.key = nn.Linear(config.encoder_width, self.all_head_size)
+            self.value = nn.Linear(config.encoder_width, self.all_head_size)
+        else:
+            self.key = nn.Linear(config.hidden_size, self.all_head_size)
+            self.value = nn.Linear(config.hidden_size, self.all_head_size)
+
+        self.dropout = nn.Dropout(config.attention_probs_dropout_prob)
+        self.position_embedding_type = getattr(config, "position_embedding_type", "absolute")
+        if self.position_embedding_type == "relative_key" or self.position_embedding_type == "relative_key_query":
+            self.max_position_embeddings = config.max_position_embeddings
+            self.distance_embedding = nn.Embedding(2 * config.max_position_embeddings - 1, self.attention_head_size)
+        self.save_attention = False   
+            
+    def save_attn_gradients(self, attn_gradients):
+        self.attn_gradients = attn_gradients
+        
+    def get_attn_gradients(self):
+        return self.attn_gradients
+    
+    def save_attention_map(self, attention_map):
+        self.attention_map = attention_map
+        
+    def get_attention_map(self):
+        return self.attention_map
+    
+    def transpose_for_scores(self, x):
+        new_x_shape = x.size()[:-1] + (self.num_attention_heads, self.attention_head_size)
+        x = x.view(*new_x_shape)
+        return x.permute(0, 2, 1, 3)
+
+    def forward(
+        self,
+        hidden_states,
+        attention_mask=None,
+        head_mask=None,
+        encoder_hidden_states=None,
+        encoder_attention_mask=None,
+        past_key_value=None,
+        output_attentions=False,
+    ):
+        mixed_query_layer = self.query(hidden_states)
+
+        # If this is instantiated as a cross-attention module, the keys
+        # and values come from an encoder; the attention mask needs to be
+        # such that the encoder's padding tokens are not attended to.
+        is_cross_attention = encoder_hidden_states is not None
+
+        if is_cross_attention:
+            key_layer = self.transpose_for_scores(self.key(encoder_hidden_states))
+            value_layer = self.transpose_for_scores(self.value(encoder_hidden_states))
+            attention_mask = encoder_attention_mask
+        elif past_key_value is not None:
+            key_layer = self.transpose_for_scores(self.key(hidden_states))
+            value_layer = self.transpose_for_scores(self.value(hidden_states))
+            key_layer = torch.cat([past_key_value[0], key_layer], dim=2)
+            value_layer = torch.cat([past_key_value[1], value_layer], dim=2)
+        else:
+            key_layer = self.transpose_for_scores(self.key(hidden_states))
+            value_layer = self.transpose_for_scores(self.value(hidden_states))
+
+        query_layer = self.transpose_for_scores(mixed_query_layer)
+
+        past_key_value = (key_layer, value_layer)
+
+        # Take the dot product between "query" and "key" to get the raw attention scores.
+        attention_scores = torch.matmul(query_layer, key_layer.transpose(-1, -2))
+
+        if self.position_embedding_type == "relative_key" or self.position_embedding_type == "relative_key_query":
+            seq_length = hidden_states.size()[1]
+            position_ids_l = torch.arange(seq_length, dtype=torch.long, device=hidden_states.device).view(-1, 1)
+            position_ids_r = torch.arange(seq_length, dtype=torch.long, device=hidden_states.device).view(1, -1)
+            distance = position_ids_l - position_ids_r
+            positional_embedding = self.distance_embedding(distance + self.max_position_embeddings - 1)
+            positional_embedding = positional_embedding.to(dtype=query_layer.dtype)  # fp16 compatibility
+
+            if self.position_embedding_type == "relative_key":
+                relative_position_scores = torch.einsum("bhld,lrd->bhlr", query_layer, positional_embedding)
+                attention_scores = attention_scores + relative_position_scores
+            elif self.position_embedding_type == "relative_key_query":
+                relative_position_scores_query = torch.einsum("bhld,lrd->bhlr", query_layer, positional_embedding)
+                relative_position_scores_key = torch.einsum("bhrd,lrd->bhlr", key_layer, positional_embedding)
+                attention_scores = attention_scores + relative_position_scores_query + relative_position_scores_key
+
+        attention_scores = attention_scores / math.sqrt(self.attention_head_size)
+        if attention_mask is not None:
+            # Apply the attention mask is (precomputed for all layers in BertModel forward() function)
+            attention_scores = attention_scores + attention_mask
+
+        # Normalize the attention scores to probabilities.
+        attention_probs = nn.Softmax(dim=-1)(attention_scores)
+        
+        if is_cross_attention and self.save_attention:
+            self.save_attention_map(attention_probs)
+            attention_probs.register_hook(self.save_attn_gradients)         
+
+        # This is actually dropping out entire tokens to attend to, which might
+        # seem a bit unusual, but is taken from the original Transformer paper.
+        attention_probs_dropped = self.dropout(attention_probs)
+
+        # Mask heads if we want to
+        if head_mask is not None:
+            attention_probs_dropped = attention_probs_dropped * head_mask
+
+        context_layer = torch.matmul(attention_probs_dropped, value_layer)
+
+        context_layer = context_layer.permute(0, 2, 1, 3).contiguous()
+        new_context_layer_shape = context_layer.size()[:-2] + (self.all_head_size,)
+        context_layer = context_layer.view(*new_context_layer_shape)
+
+        outputs = (context_layer, attention_probs) if output_attentions else (context_layer,)
+
+        outputs = outputs + (past_key_value,)
+        return outputs
+
+
+class BertSelfOutput(nn.Module):
+    def __init__(self, config, twin=False, merge=False):     
+        super().__init__()
+        self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
+        self.dropout = nn.Dropout(config.hidden_dropout_prob)        
+        if twin:
+            self.dense0 = nn.Linear(config.hidden_size, config.hidden_size)
+            self.dense1 = nn.Linear(config.hidden_size, config.hidden_size)         
+        else:
+            self.dense = nn.Linear(config.hidden_size, config.hidden_size)
+        if merge:
+            self.act =  ACT2FN[config.hidden_act]
+            self.merge_layer = nn.Linear(config.hidden_size * 2, config.hidden_size)
+            self.merge = True
+        else:
+            self.merge = False
+
+    def forward(self, hidden_states, input_tensor):
+        if type(hidden_states) == list:
+            hidden_states0 = self.dense0(hidden_states[0])
+            hidden_states1 = self.dense1(hidden_states[1])        
+            if self.merge:  
+                #hidden_states = self.merge_layer(self.act(torch.cat([hidden_states0,hidden_states1],dim=-1)))
+                hidden_states = self.merge_layer(torch.cat([hidden_states0,hidden_states1],dim=-1))
+            else:
+                hidden_states = (hidden_states0+hidden_states1)/2
+        else:    
+            hidden_states = self.dense(hidden_states)
+        hidden_states = self.dropout(hidden_states)
+        hidden_states = self.LayerNorm(hidden_states + input_tensor)
+        return hidden_states
+
+
+class BertAttention(nn.Module):
+    def __init__(self, config, is_cross_attention=False, layer_num=-1):
+        super().__init__()
+        if is_cross_attention:
+            self.self0 = BertSelfAttention(config, is_cross_attention)
+            self.self1 = BertSelfAttention(config, is_cross_attention)
+        else:    
+            self.self = BertSelfAttention(config, is_cross_attention)
+        self.output = BertSelfOutput(config, twin=is_cross_attention, merge=(is_cross_attention and layer_num>=6))
+        self.pruned_heads = set()
+
+    def prune_heads(self, heads):
+        if len(heads) == 0:
+            return
+        heads, index = find_pruneable_heads_and_indices(
+            heads, self.self.num_attention_heads, self.self.attention_head_size, self.pruned_heads
+        )
+
+        # Prune linear layers
+        self.self.query = prune_linear_layer(self.self.query, index)
+        self.self.key = prune_linear_layer(self.self.key, index)
+        self.self.value = prune_linear_layer(self.self.value, index)
+        self.output.dense = prune_linear_layer(self.output.dense, index, dim=1)
+
+        # Update hyper params and store pruned heads
+        self.self.num_attention_heads = self.self.num_attention_heads - len(heads)
+        self.self.all_head_size = self.self.attention_head_size * self.self.num_attention_heads
+        self.pruned_heads = self.pruned_heads.union(heads)
+
+    def forward(
+        self,
+        hidden_states,
+        attention_mask=None,
+        head_mask=None,
+        encoder_hidden_states=None,
+        encoder_attention_mask=None,
+        past_key_value=None,
+        output_attentions=False,
+    ):        
+        if type(encoder_hidden_states)==list:   
+            self_outputs0 = self.self0(
+                hidden_states,
+                attention_mask,
+                head_mask,
+                encoder_hidden_states[0],
+                encoder_attention_mask[0],
+                past_key_value,
+                output_attentions,
+            )
+            self_outputs1 = self.self1(
+                hidden_states,
+                attention_mask,
+                head_mask,
+                encoder_hidden_states[1],
+                encoder_attention_mask[1],
+                past_key_value,
+                output_attentions,
+            )                        
+            attention_output = self.output([self_outputs0[0],self_outputs1[0]], hidden_states)
+    
+            outputs = (attention_output,) + self_outputs0[1:]  # add attentions if we output them
+        else:        
+            self_outputs = self.self(
+                hidden_states,
+                attention_mask,
+                head_mask,
+                encoder_hidden_states,
+                encoder_attention_mask,
+                past_key_value,
+                output_attentions,
+            )
+            attention_output = self.output(self_outputs[0], hidden_states)
+            outputs = (attention_output,) + self_outputs[1:]  # add attentions if we output them
+        return outputs
+
+
+class BertIntermediate(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.dense = nn.Linear(config.hidden_size, config.intermediate_size)
+        if isinstance(config.hidden_act, str):
+            self.intermediate_act_fn = ACT2FN[config.hidden_act]
+        else:
+            self.intermediate_act_fn = config.hidden_act
+
+    def forward(self, hidden_states):
+        hidden_states = self.dense(hidden_states)
+        hidden_states = self.intermediate_act_fn(hidden_states)
+        return hidden_states
+
+
+class BertOutput(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.dense = nn.Linear(config.intermediate_size, config.hidden_size)
+        self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
+        self.dropout = nn.Dropout(config.hidden_dropout_prob)
+
+    def forward(self, hidden_states, input_tensor):
+        hidden_states = self.dense(hidden_states)
+        hidden_states = self.dropout(hidden_states)
+        hidden_states = self.LayerNorm(hidden_states + input_tensor)
+        return hidden_states
+
+
+class BertLayer(nn.Module):
+    def __init__(self, config, layer_num):
+        super().__init__()
+        self.config = config
+        self.chunk_size_feed_forward = config.chunk_size_feed_forward
+        self.seq_len_dim = 1
+        self.attention = BertAttention(config)      
+        self.layer_num = layer_num          
+        if self.config.add_cross_attention:
+            self.crossattention = BertAttention(config, is_cross_attention=self.config.add_cross_attention, layer_num=layer_num)
+        self.intermediate = BertIntermediate(config)
+        self.output = BertOutput(config)
+
+    def forward(
+        self,
+        hidden_states,
+        attention_mask=None,
+        head_mask=None,
+        encoder_hidden_states=None,
+        encoder_attention_mask=None,
+        past_key_value=None,
+        output_attentions=False,
+        mode=None,
+    ):
+        # decoder uni-directional self-attention cached key/values tuple is at positions 1,2
+        self_attn_past_key_value = past_key_value[:2] if past_key_value is not None else None
+        self_attention_outputs = self.attention(
+            hidden_states,
+            attention_mask,
+            head_mask,
+            output_attentions=output_attentions,
+            past_key_value=self_attn_past_key_value,
+        )
+        attention_output = self_attention_outputs[0]
+
+        outputs = self_attention_outputs[1:-1]
+        present_key_value = self_attention_outputs[-1]
+
+        if mode=='multimodal':
+            assert encoder_hidden_states is not None, "encoder_hidden_states must be given for cross-attention layers"
+            cross_attention_outputs = self.crossattention(
+                attention_output,
+                attention_mask,
+                head_mask,
+                encoder_hidden_states,
+                encoder_attention_mask,
+                output_attentions=output_attentions,
+            )
+            attention_output = cross_attention_outputs[0]
+            outputs = outputs + cross_attention_outputs[1:-1]  # add cross attentions if we output attention weights                               
+        layer_output = apply_chunking_to_forward(
+            self.feed_forward_chunk, self.chunk_size_feed_forward, self.seq_len_dim, attention_output
+        )
+        outputs = (layer_output,) + outputs
+
+        outputs = outputs + (present_key_value,)
+
+        return outputs
+
+    def feed_forward_chunk(self, attention_output):
+        intermediate_output = self.intermediate(attention_output)
+        layer_output = self.output(intermediate_output, attention_output)
+        return layer_output
+
+
+class BertEncoder(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.config = config
+        self.layer = nn.ModuleList([BertLayer(config,i) for i in range(config.num_hidden_layers)])
+        self.gradient_checkpointing = False
+
+    def forward(
+        self,
+        hidden_states,
+        attention_mask=None,
+        head_mask=None,
+        encoder_hidden_states=None,
+        encoder_attention_mask=None,
+        past_key_values=None,
+        use_cache=None,
+        output_attentions=False,
+        output_hidden_states=False,
+        return_dict=True,
+        mode='multimodal',
+    ):
+        all_hidden_states = () if output_hidden_states else None
+        all_self_attentions = () if output_attentions else None
+        all_cross_attentions = () if output_attentions and self.config.add_cross_attention else None
+
+        next_decoder_cache = () if use_cache else None
+               
+        for i in range(self.config.num_hidden_layers):
+            layer_module = self.layer[i]
+            if output_hidden_states:
+                all_hidden_states = all_hidden_states + (hidden_states,)
+
+            layer_head_mask = head_mask[i] if head_mask is not None else None
+            past_key_value = past_key_values[i] if past_key_values is not None else None
+
+            if self.gradient_checkpointing and self.training:
+
+                if use_cache:
+                    logger.warn(
+                        "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
+                    )
+                    use_cache = False
+
+                def create_custom_forward(module):
+                    def custom_forward(*inputs):
+                        return module(*inputs, past_key_value, output_attentions)
+
+                    return custom_forward
+
+                layer_outputs = torch.utils.checkpoint.checkpoint(
+                    create_custom_forward(layer_module),
+                    hidden_states,
+                    attention_mask,
+                    layer_head_mask,
+                    encoder_hidden_states,
+                    encoder_attention_mask,
+                    mode=mode,
+                )
+            else:
+                layer_outputs = layer_module(
+                    hidden_states,
+                    attention_mask,
+                    layer_head_mask,
+                    encoder_hidden_states,
+                    encoder_attention_mask,
+                    past_key_value,
+                    output_attentions,
+                    mode=mode,
+                )
+
+            hidden_states = layer_outputs[0]
+            if use_cache:
+                next_decoder_cache += (layer_outputs[-1],)
+            if output_attentions:
+                all_self_attentions = all_self_attentions + (layer_outputs[1],)
+
+        if output_hidden_states:
+            all_hidden_states = all_hidden_states + (hidden_states,)
+
+        if not return_dict:
+            return tuple(
+                v
+                for v in [
+                    hidden_states,
+                    next_decoder_cache,
+                    all_hidden_states,
+                    all_self_attentions,
+                    all_cross_attentions,
+                ]
+                if v is not None
+            )
+        return BaseModelOutputWithPastAndCrossAttentions(
+            last_hidden_state=hidden_states,
+            past_key_values=next_decoder_cache,
+            hidden_states=all_hidden_states,
+            attentions=all_self_attentions,
+            cross_attentions=all_cross_attentions,
+        )
+
+
+class BertPooler(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.dense = nn.Linear(config.hidden_size, config.hidden_size)
+        self.activation = nn.Tanh()
+
+    def forward(self, hidden_states):
+        # We "pool" the model by simply taking the hidden state corresponding
+        # to the first token.
+        first_token_tensor = hidden_states[:, 0]
+        pooled_output = self.dense(first_token_tensor)
+        pooled_output = self.activation(pooled_output)
+        return pooled_output
+
+
+class BertPredictionHeadTransform(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.dense = nn.Linear(config.hidden_size, config.hidden_size)
+        if isinstance(config.hidden_act, str):
+            self.transform_act_fn = ACT2FN[config.hidden_act]
+        else:
+            self.transform_act_fn = config.hidden_act
+        self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
+
+    def forward(self, hidden_states):
+        hidden_states = self.dense(hidden_states)
+        hidden_states = self.transform_act_fn(hidden_states)
+        hidden_states = self.LayerNorm(hidden_states)
+        return hidden_states
+
+
+class BertLMPredictionHead(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.transform = BertPredictionHeadTransform(config)
+
+        # The output weights are the same as the input embeddings, but there is
+        # an output-only bias for each token.
+        self.decoder = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
+
+        self.bias = nn.Parameter(torch.zeros(config.vocab_size))
+
+        # Need a link between the two variables so that the bias is correctly resized with `resize_token_embeddings`
+        self.decoder.bias = self.bias
+
+    def forward(self, hidden_states):
+        hidden_states = self.transform(hidden_states)
+        hidden_states = self.decoder(hidden_states)
+        return hidden_states
+
+
+class BertOnlyMLMHead(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.predictions = BertLMPredictionHead(config)
+
+    def forward(self, sequence_output):
+        prediction_scores = self.predictions(sequence_output)
+        return prediction_scores
+
+
+class BertPreTrainedModel(PreTrainedModel):
+    """
+    An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained
+    models.
+    """
+
+    config_class = BertConfig
+    base_model_prefix = "bert"
+    _keys_to_ignore_on_load_missing = [r"position_ids"]
+
+    def _init_weights(self, module):
+        """ Initialize the weights """
+        if isinstance(module, (nn.Linear, nn.Embedding)):
+            # Slightly different from the TF version which uses truncated_normal for initialization
+            # cf https://github.com/pytorch/pytorch/pull/5617
+            module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)
+        elif isinstance(module, nn.LayerNorm):
+            module.bias.data.zero_()
+            module.weight.data.fill_(1.0)
+        if isinstance(module, nn.Linear) and module.bias is not None:
+            module.bias.data.zero_()
+
+
+class BertModel(BertPreTrainedModel):
+    """
+    The model can behave as an encoder (with only self-attention) as well as a decoder, in which case a layer of
+    cross-attention is added between the self-attention layers, following the architecture described in `Attention is
+    all you need <https://arxiv.org/abs/1706.03762>`__ by Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit,
+    Llion Jones, Aidan N. Gomez, Lukasz Kaiser and Illia Polosukhin.
+    argument and :obj:`add_cross_attention` set to :obj:`True`; an :obj:`encoder_hidden_states` is then expected as an
+    input to the forward pass.
+    """
+
+    def __init__(self, config, add_pooling_layer=True):
+        super().__init__(config)
+        self.config = config
+
+        self.embeddings = BertEmbeddings(config)
+        
+        self.encoder = BertEncoder(config)
+
+        self.pooler = BertPooler(config) if add_pooling_layer else None
+
+        self.init_weights()
+ 
+
+    def get_input_embeddings(self):
+        return self.embeddings.word_embeddings
+
+    def set_input_embeddings(self, value):
+        self.embeddings.word_embeddings = value
+
+    def _prune_heads(self, heads_to_prune):
+        """
+        Prunes heads of the model. heads_to_prune: dict of {layer_num: list of heads to prune in this layer} See base
+        class PreTrainedModel
+        """
+        for layer, heads in heads_to_prune.items():
+            self.encoder.layer[layer].attention.prune_heads(heads)
+
+    
+    def get_extended_attention_mask(self, attention_mask: Tensor, input_shape: Tuple[int], device: device, is_decoder: bool) -> Tensor:
+        """
+        Makes broadcastable attention and causal masks so that future and masked tokens are ignored.
+
+        Arguments:
+            attention_mask (:obj:`torch.Tensor`):
+                Mask with ones indicating tokens to attend to, zeros for tokens to ignore.
+            input_shape (:obj:`Tuple[int]`):
+                The shape of the input to the model.
+            device: (:obj:`torch.device`):
+                The device of the input to the model.
+
+        Returns:
+            :obj:`torch.Tensor` The extended attention mask, with a the same dtype as :obj:`attention_mask.dtype`.
+        """
+        # We can provide a self-attention mask of dimensions [batch_size, from_seq_length, to_seq_length]
+        # ourselves in which case we just need to make it broadcastable to all heads.
+        if attention_mask.dim() == 3:
+            extended_attention_mask = attention_mask[:, None, :, :]
+        elif attention_mask.dim() == 2:
+            # Provided a padding mask of dimensions [batch_size, seq_length]
+            # - if the model is a decoder, apply a causal mask in addition to the padding mask
+            # - if the model is an encoder, make the mask broadcastable to [batch_size, num_heads, seq_length, seq_length]
+            if is_decoder:
+                batch_size, seq_length = input_shape
+
+                seq_ids = torch.arange(seq_length, device=device)
+                causal_mask = seq_ids[None, None, :].repeat(batch_size, seq_length, 1) <= seq_ids[None, :, None]
+                # in case past_key_values are used we need to add a prefix ones mask to the causal mask
+                # causal and attention masks must have same type with pytorch version < 1.3
+                causal_mask = causal_mask.to(attention_mask.dtype)
+   
+                if causal_mask.shape[1] < attention_mask.shape[1]:
+                    prefix_seq_len = attention_mask.shape[1] - causal_mask.shape[1]
+                    causal_mask = torch.cat(
+                        [
+                            torch.ones((batch_size, seq_length, prefix_seq_len), device=device, dtype=causal_mask.dtype),
+                            causal_mask,
+                        ],
+                        axis=-1,
+                    )                     
+
+                extended_attention_mask = causal_mask[:, None, :, :] * attention_mask[:, None, None, :]
+            else:
+                extended_attention_mask = attention_mask[:, None, None, :]
+        else:
+            raise ValueError(
+                "Wrong shape for input_ids (shape {}) or attention_mask (shape {})".format(
+                    input_shape, attention_mask.shape
+                )
+            )
+
+        # Since attention_mask is 1.0 for positions we want to attend and 0.0 for
+        # masked positions, this operation will create a tensor which is 0.0 for
+        # positions we want to attend and -10000.0 for masked positions.
+        # Since we are adding it to the raw scores before the softmax, this is
+        # effectively the same as removing these entirely.
+        extended_attention_mask = extended_attention_mask.to(dtype=self.dtype)  # fp16 compatibility
+        extended_attention_mask = (1.0 - extended_attention_mask) * -10000.0
+        return extended_attention_mask
+    
+    def forward(
+        self,
+        input_ids=None,
+        attention_mask=None,
+        position_ids=None,
+        head_mask=None,
+        inputs_embeds=None,
+        encoder_embeds=None,
+        encoder_hidden_states=None,
+        encoder_attention_mask=None,
+        past_key_values=None,
+        use_cache=None,
+        output_attentions=None,
+        output_hidden_states=None,
+        return_dict=None,
+        is_decoder=False,
+        mode='multimodal',
+    ):
+        r"""
+        encoder_hidden_states  (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length, hidden_size)`, `optional`):
+            Sequence of hidden-states at the output of the last layer of the encoder. Used in the cross-attention if
+            the model is configured as a decoder.
+        encoder_attention_mask (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`):
+            Mask to avoid performing attention on the padding token indices of the encoder input. This mask is used in
+            the cross-attention if the model is configured as a decoder. Mask values selected in ``[0, 1]``:
+            - 1 for tokens that are **not masked**,
+            - 0 for tokens that are **masked**.
+        past_key_values (:obj:`tuple(tuple(torch.FloatTensor))` of length :obj:`config.n_layers` with each tuple having 4 tensors of shape :obj:`(batch_size, num_heads, sequence_length - 1, embed_size_per_head)`):
+            Contains precomputed key and value hidden states of the attention blocks. Can be used to speed up decoding.
+            If :obj:`past_key_values` are used, the user can optionally input only the last :obj:`decoder_input_ids`
+            (those that don't have their past key value states given to this model) of shape :obj:`(batch_size, 1)`
+            instead of all :obj:`decoder_input_ids` of shape :obj:`(batch_size, sequence_length)`.
+        use_cache (:obj:`bool`, `optional`):
+            If set to :obj:`True`, :obj:`past_key_values` key value states are returned and can be used to speed up
+            decoding (see :obj:`past_key_values`).
+        """
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        output_hidden_states = (
+            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        )
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        if is_decoder:
+            use_cache = use_cache if use_cache is not None else self.config.use_cache
+        else:
+            use_cache = False
+
+        if input_ids is not None and inputs_embeds is not None:
+            raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
+        elif input_ids is not None:
+            input_shape = input_ids.size()
+            batch_size, seq_length = input_shape
+            device = input_ids.device
+        elif inputs_embeds is not None:
+            input_shape = inputs_embeds.size()[:-1]
+            batch_size, seq_length = input_shape
+            device = inputs_embeds.device
+        elif encoder_embeds is not None:    
+            input_shape = encoder_embeds.size()[:-1]
+            batch_size, seq_length = input_shape 
+            device = encoder_embeds.device
+        else:
+            raise ValueError("You have to specify either input_ids or inputs_embeds or encoder_embeds")
+
+        # past_key_values_length
+        past_key_values_length = past_key_values[0][0].shape[2] if past_key_values is not None else 0
+
+        if attention_mask is None:
+            attention_mask = torch.ones(((batch_size, seq_length + past_key_values_length)), device=device)
+            
+        # We can provide a self-attention mask of dimensions [batch_size, from_seq_length, to_seq_length]
+        # ourselves in which case we just need to make it broadcastable to all heads.
+        extended_attention_mask: torch.Tensor = self.get_extended_attention_mask(attention_mask, input_shape, 
+                                                                                 device, is_decoder)
+
+        # If a 2D or 3D attention mask is provided for the cross-attention
+        # we need to make broadcastable to [batch_size, num_heads, seq_length, seq_length]
+        if encoder_hidden_states is not None:
+            if type(encoder_hidden_states) == list:
+                encoder_batch_size, encoder_sequence_length, _ = encoder_hidden_states[0].size()
+            else:
+                encoder_batch_size, encoder_sequence_length, _ = encoder_hidden_states.size()
+            encoder_hidden_shape = (encoder_batch_size, encoder_sequence_length)
+            
+            if type(encoder_attention_mask) == list:
+                encoder_extended_attention_mask = [self.invert_attention_mask(mask) for mask in encoder_attention_mask]
+            elif encoder_attention_mask is None:
+                encoder_attention_mask = torch.ones(encoder_hidden_shape, device=device)
+                encoder_extended_attention_mask = self.invert_attention_mask(encoder_attention_mask)
+            else:    
+                encoder_extended_attention_mask = self.invert_attention_mask(encoder_attention_mask)
+        else:
+            encoder_extended_attention_mask = None
+
+        # Prepare head mask if needed
+        # 1.0 in head_mask indicate we keep the head
+        # attention_probs has shape bsz x n_heads x N x N
+        # input head_mask has shape [num_heads] or [num_hidden_layers x num_heads]
+        # and head_mask is converted to shape [num_hidden_layers x batch x num_heads x seq_length x seq_length]
+        head_mask = self.get_head_mask(head_mask, self.config.num_hidden_layers)
+        
+        if encoder_embeds is None:
+            embedding_output = self.embeddings(
+                input_ids=input_ids,
+                position_ids=position_ids,
+                inputs_embeds=inputs_embeds,
+                past_key_values_length=past_key_values_length,
+            )
+        else:
+            embedding_output = encoder_embeds
+            
+        encoder_outputs = self.encoder(
+            embedding_output,
+            attention_mask=extended_attention_mask,
+            head_mask=head_mask,
+            encoder_hidden_states=encoder_hidden_states,
+            encoder_attention_mask=encoder_extended_attention_mask,
+            past_key_values=past_key_values,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+            mode=mode,
+        )
+        sequence_output = encoder_outputs[0]
+        pooled_output = self.pooler(sequence_output) if self.pooler is not None else None
+
+        if not return_dict:
+            return (sequence_output, pooled_output) + encoder_outputs[1:]
+
+        return BaseModelOutputWithPoolingAndCrossAttentions(
+            last_hidden_state=sequence_output,
+            pooler_output=pooled_output,
+            past_key_values=encoder_outputs.past_key_values,
+            hidden_states=encoder_outputs.hidden_states,
+            attentions=encoder_outputs.attentions,
+            cross_attentions=encoder_outputs.cross_attentions,
+        )
+

extras/BLIP/models/vit.py

@@ -0,0 +1,308 @@
+'''
+ * Copyright (c) 2022, salesforce.com, inc.
+ * All rights reserved.
+ * SPDX-License-Identifier: BSD-3-Clause
+ * For full license text, see LICENSE.txt file in the repo root or https://opensource.org/licenses/BSD-3-Clause
+ * By Junnan Li
+ * Based on timm code base
+ * https://github.com/rwightman/pytorch-image-models/tree/master/timm
+'''
+
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+from functools import partial
+
+from timm.models.vision_transformer import _cfg, PatchEmbed
+from timm.models.registry import register_model
+from timm.models.layers import trunc_normal_, DropPath
+from timm.models.helpers import named_apply, adapt_input_conv
+
+
+def checkpoint_wrapper(x):
+    return x
+
+
+class Mlp(nn.Module):
+    """ MLP as used in Vision Transformer, MLP-Mixer and related networks
+    """
+    def __init__(self, in_features, hidden_features=None, out_features=None, act_layer=nn.GELU, drop=0.):
+        super().__init__()
+        out_features = out_features or in_features
+        hidden_features = hidden_features or in_features
+        self.fc1 = nn.Linear(in_features, hidden_features)
+        self.act = act_layer()
+        self.fc2 = nn.Linear(hidden_features, out_features)
+        self.drop = nn.Dropout(drop)
+
+    def forward(self, x):
+        x = self.fc1(x)
+        x = self.act(x)
+        x = self.drop(x)
+        x = self.fc2(x)
+        x = self.drop(x)
+        return x
+
+
+class Attention(nn.Module):
+    def __init__(self, dim, num_heads=8, qkv_bias=False, qk_scale=None, attn_drop=0., proj_drop=0.):
+        super().__init__()
+        self.num_heads = num_heads
+        head_dim = dim // num_heads
+        # NOTE scale factor was wrong in my original version, can set manually to be compat with prev weights
+        self.scale = qk_scale or head_dim ** -0.5
+        self.qkv = nn.Linear(dim, dim * 3, bias=qkv_bias)
+        self.attn_drop = nn.Dropout(attn_drop)
+        self.proj = nn.Linear(dim, dim)
+        self.proj_drop = nn.Dropout(proj_drop)
+        self.attn_gradients = None
+        self.attention_map = None
+        
+    def save_attn_gradients(self, attn_gradients):
+        self.attn_gradients = attn_gradients
+        
+    def get_attn_gradients(self):
+        return self.attn_gradients
+    
+    def save_attention_map(self, attention_map):
+        self.attention_map = attention_map
+        
+    def get_attention_map(self):
+        return self.attention_map
+    
+    def forward(self, x, register_hook=False):
+        B, N, C = x.shape
+        qkv = self.qkv(x).reshape(B, N, 3, self.num_heads, C // self.num_heads).permute(2, 0, 3, 1, 4)
+        q, k, v = qkv[0], qkv[1], qkv[2]   # make torchscript happy (cannot use tensor as tuple)
+
+        attn = (q @ k.transpose(-2, -1)) * self.scale
+        attn = attn.softmax(dim=-1)
+        attn = self.attn_drop(attn)
+                
+        if register_hook:
+            self.save_attention_map(attn)
+            attn.register_hook(self.save_attn_gradients)        
+
+        x = (attn @ v).transpose(1, 2).reshape(B, N, C)
+        x = self.proj(x)
+        x = self.proj_drop(x)
+        return x
+
+
+class Block(nn.Module):
+
+    def __init__(self, dim, num_heads, mlp_ratio=4., qkv_bias=False, qk_scale=None, drop=0., attn_drop=0.,
+                 drop_path=0., act_layer=nn.GELU, norm_layer=nn.LayerNorm, use_grad_checkpointing=False):
+        super().__init__()
+        self.norm1 = norm_layer(dim)
+        self.attn = Attention(
+            dim, num_heads=num_heads, qkv_bias=qkv_bias, qk_scale=qk_scale, attn_drop=attn_drop, proj_drop=drop)
+        # NOTE: drop path for stochastic depth, we shall see if this is better than dropout here
+        self.drop_path = DropPath(drop_path) if drop_path > 0. else nn.Identity()
+        self.norm2 = norm_layer(dim)
+        mlp_hidden_dim = int(dim * mlp_ratio)
+        self.mlp = Mlp(in_features=dim, hidden_features=mlp_hidden_dim, act_layer=act_layer, drop=drop)
+
+        if use_grad_checkpointing:
+            self.attn = checkpoint_wrapper(self.attn)
+            self.mlp = checkpoint_wrapper(self.mlp)
+
+    def forward(self, x, register_hook=False):
+        x = x + self.drop_path(self.attn(self.norm1(x), register_hook=register_hook))
+        x = x + self.drop_path(self.mlp(self.norm2(x)))
+        return x
+
+    
+class VisionTransformer(nn.Module):
+    """ Vision Transformer
+    A PyTorch impl of : `An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale`  -
+        https://arxiv.org/abs/2010.11929
+    """
+    def __init__(self, img_size=224, patch_size=16, in_chans=3, num_classes=1000, embed_dim=768, depth=12,
+                 num_heads=12, mlp_ratio=4., qkv_bias=True, qk_scale=None, representation_size=None,
+                 drop_rate=0., attn_drop_rate=0., drop_path_rate=0., norm_layer=None, 
+                 use_grad_checkpointing=False, ckpt_layer=0):
+        """
+        Args:
+            img_size (int, tuple): input image size
+            patch_size (int, tuple): patch size
+            in_chans (int): number of input channels
+            num_classes (int): number of classes for classification head
+            embed_dim (int): embedding dimension
+            depth (int): depth of transformer
+            num_heads (int): number of attention heads
+            mlp_ratio (int): ratio of mlp hidden dim to embedding dim
+            qkv_bias (bool): enable bias for qkv if True
+            qk_scale (float): override default qk scale of head_dim ** -0.5 if set
+            representation_size (Optional[int]): enable and set representation layer (pre-logits) to this value if set
+            drop_rate (float): dropout rate
+            attn_drop_rate (float): attention dropout rate
+            drop_path_rate (float): stochastic depth rate
+            norm_layer: (nn.Module): normalization layer
+        """
+        super().__init__()
+        self.num_features = self.embed_dim = embed_dim  # num_features for consistency with other models
+        norm_layer = norm_layer or partial(nn.LayerNorm, eps=1e-6)
+
+        self.patch_embed = PatchEmbed(
+            img_size=img_size, patch_size=patch_size, in_chans=in_chans, embed_dim=embed_dim)
+
+        num_patches = self.patch_embed.num_patches
+
+        self.cls_token = nn.Parameter(torch.zeros(1, 1, embed_dim))
+        self.pos_embed = nn.Parameter(torch.zeros(1, num_patches + 1, embed_dim))
+        self.pos_drop = nn.Dropout(p=drop_rate)
+
+        dpr = [x.item() for x in torch.linspace(0, drop_path_rate, depth)]  # stochastic depth decay rule
+        self.blocks = nn.ModuleList([
+            Block(
+                dim=embed_dim, num_heads=num_heads, mlp_ratio=mlp_ratio, qkv_bias=qkv_bias, qk_scale=qk_scale,
+                drop=drop_rate, attn_drop=attn_drop_rate, drop_path=dpr[i], norm_layer=norm_layer,
+                use_grad_checkpointing=(use_grad_checkpointing and i>=depth-ckpt_layer)
+            )
+            for i in range(depth)])
+        self.norm = norm_layer(embed_dim)
+
+        trunc_normal_(self.pos_embed, std=.02)
+        trunc_normal_(self.cls_token, std=.02)
+        self.apply(self._init_weights)
+
+    def _init_weights(self, m):
+        if isinstance(m, nn.Linear):
+            trunc_normal_(m.weight, std=.02)
+            if isinstance(m, nn.Linear) and m.bias is not None:
+                nn.init.constant_(m.bias, 0)
+        elif isinstance(m, nn.LayerNorm):
+            nn.init.constant_(m.bias, 0)
+            nn.init.constant_(m.weight, 1.0)
+
+    @torch.jit.ignore
+    def no_weight_decay(self):
+        return {'pos_embed', 'cls_token'}
+
+    def forward(self, x, register_blk=-1):
+        B = x.shape[0]
+        x = self.patch_embed(x)
+
+        cls_tokens = self.cls_token.expand(B, -1, -1)  # stole cls_tokens impl from Phil Wang, thanks
+        x = torch.cat((cls_tokens, x), dim=1)
+  
+        x = x + self.pos_embed[:,:x.size(1),:]
+        x = self.pos_drop(x)
+
+        for i,blk in enumerate(self.blocks):
+            x = blk(x, register_blk==i)
+        x = self.norm(x)
+        
+        return x
+
+    @torch.jit.ignore()
+    def load_pretrained(self, checkpoint_path, prefix=''):
+        _load_weights(self, checkpoint_path, prefix)
+        
+
+@torch.no_grad()
+def _load_weights(model: VisionTransformer, checkpoint_path: str, prefix: str = ''):
+    """ Load weights from .npz checkpoints for official Google Brain Flax implementation
+    """
+    import numpy as np
+
+    def _n2p(w, t=True):
+        if w.ndim == 4 and w.shape[0] == w.shape[1] == w.shape[2] == 1:
+            w = w.flatten()
+        if t:
+            if w.ndim == 4:
+                w = w.transpose([3, 2, 0, 1])
+            elif w.ndim == 3:
+                w = w.transpose([2, 0, 1])
+            elif w.ndim == 2:
+                w = w.transpose([1, 0])
+        return torch.from_numpy(w)
+
+    w = np.load(checkpoint_path)
+    if not prefix and 'opt/target/embedding/kernel' in w:
+        prefix = 'opt/target/'
+
+    if hasattr(model.patch_embed, 'backbone'):
+        # hybrid
+        backbone = model.patch_embed.backbone
+        stem_only = not hasattr(backbone, 'stem')
+        stem = backbone if stem_only else backbone.stem
+        stem.conv.weight.copy_(adapt_input_conv(stem.conv.weight.shape[1], _n2p(w[f'{prefix}conv_root/kernel'])))
+        stem.norm.weight.copy_(_n2p(w[f'{prefix}gn_root/scale']))
+        stem.norm.bias.copy_(_n2p(w[f'{prefix}gn_root/bias']))
+        if not stem_only:
+            for i, stage in enumerate(backbone.stages):
+                for j, block in enumerate(stage.blocks):
+                    bp = f'{prefix}block{i + 1}/unit{j + 1}/'
+                    for r in range(3):
+                        getattr(block, f'conv{r + 1}').weight.copy_(_n2p(w[f'{bp}conv{r + 1}/kernel']))
+                        getattr(block, f'norm{r + 1}').weight.copy_(_n2p(w[f'{bp}gn{r + 1}/scale']))
+                        getattr(block, f'norm{r + 1}').bias.copy_(_n2p(w[f'{bp}gn{r + 1}/bias']))
+                    if block.downsample is not None:
+                        block.downsample.conv.weight.copy_(_n2p(w[f'{bp}conv_proj/kernel']))
+                        block.downsample.norm.weight.copy_(_n2p(w[f'{bp}gn_proj/scale']))
+                        block.downsample.norm.bias.copy_(_n2p(w[f'{bp}gn_proj/bias']))
+        embed_conv_w = _n2p(w[f'{prefix}embedding/kernel'])
+    else:
+        embed_conv_w = adapt_input_conv(
+            model.patch_embed.proj.weight.shape[1], _n2p(w[f'{prefix}embedding/kernel']))
+    model.patch_embed.proj.weight.copy_(embed_conv_w)
+    model.patch_embed.proj.bias.copy_(_n2p(w[f'{prefix}embedding/bias']))
+    model.cls_token.copy_(_n2p(w[f'{prefix}cls'], t=False))
+    pos_embed_w = _n2p(w[f'{prefix}Transformer/posembed_input/pos_embedding'], t=False)
+    if pos_embed_w.shape != model.pos_embed.shape:
+        pos_embed_w = resize_pos_embed(  # resize pos embedding when different size from pretrained weights
+            pos_embed_w, model.pos_embed, getattr(model, 'num_tokens', 1), model.patch_embed.grid_size)
+    model.pos_embed.copy_(pos_embed_w)
+    model.norm.weight.copy_(_n2p(w[f'{prefix}Transformer/encoder_norm/scale']))
+    model.norm.bias.copy_(_n2p(w[f'{prefix}Transformer/encoder_norm/bias']))
+#     if isinstance(model.head, nn.Linear) and model.head.bias.shape[0] == w[f'{prefix}head/bias'].shape[-1]:
+#         model.head.weight.copy_(_n2p(w[f'{prefix}head/kernel']))
+#         model.head.bias.copy_(_n2p(w[f'{prefix}head/bias']))
+#     if isinstance(getattr(model.pre_logits, 'fc', None), nn.Linear) and f'{prefix}pre_logits/bias' in w:
+#         model.pre_logits.fc.weight.copy_(_n2p(w[f'{prefix}pre_logits/kernel']))
+#         model.pre_logits.fc.bias.copy_(_n2p(w[f'{prefix}pre_logits/bias']))
+    for i, block in enumerate(model.blocks.children()):
+        block_prefix = f'{prefix}Transformer/encoderblock_{i}/'
+        mha_prefix = block_prefix + 'MultiHeadDotProductAttention_1/'
+        block.norm1.weight.copy_(_n2p(w[f'{block_prefix}LayerNorm_0/scale']))
+        block.norm1.bias.copy_(_n2p(w[f'{block_prefix}LayerNorm_0/bias']))
+        block.attn.qkv.weight.copy_(torch.cat([
+            _n2p(w[f'{mha_prefix}{n}/kernel'], t=False).flatten(1).T for n in ('query', 'key', 'value')]))
+        block.attn.qkv.bias.copy_(torch.cat([
+            _n2p(w[f'{mha_prefix}{n}/bias'], t=False).reshape(-1) for n in ('query', 'key', 'value')]))
+        block.attn.proj.weight.copy_(_n2p(w[f'{mha_prefix}out/kernel']).flatten(1))
+        block.attn.proj.bias.copy_(_n2p(w[f'{mha_prefix}out/bias']))
+        for r in range(2):
+            getattr(block.mlp, f'fc{r + 1}').weight.copy_(_n2p(w[f'{block_prefix}MlpBlock_3/Dense_{r}/kernel']))
+            getattr(block.mlp, f'fc{r + 1}').bias.copy_(_n2p(w[f'{block_prefix}MlpBlock_3/Dense_{r}/bias']))
+        block.norm2.weight.copy_(_n2p(w[f'{block_prefix}LayerNorm_2/scale']))
+        block.norm2.bias.copy_(_n2p(w[f'{block_prefix}LayerNorm_2/bias']))
+
+            
+def interpolate_pos_embed(pos_embed_checkpoint, visual_encoder):        
+    # interpolate position embedding
+    embedding_size = pos_embed_checkpoint.shape[-1]
+    num_patches = visual_encoder.patch_embed.num_patches
+    num_extra_tokens = visual_encoder.pos_embed.shape[-2] - num_patches
+    # height (== width) for the checkpoint position embedding
+    orig_size = int((pos_embed_checkpoint.shape[-2] - num_extra_tokens) ** 0.5)
+    # height (== width) for the new position embedding
+    new_size = int(num_patches ** 0.5)
+
+    if orig_size!=new_size:
+        # class_token and dist_token are kept unchanged
+        extra_tokens = pos_embed_checkpoint[:, :num_extra_tokens]
+        # only the position tokens are interpolated
+        pos_tokens = pos_embed_checkpoint[:, num_extra_tokens:]
+        pos_tokens = pos_tokens.reshape(-1, orig_size, orig_size, embedding_size).permute(0, 3, 1, 2)
+        pos_tokens = torch.nn.functional.interpolate(
+            pos_tokens, size=(new_size, new_size), mode='bicubic', align_corners=False)
+        pos_tokens = pos_tokens.permute(0, 2, 3, 1).flatten(1, 2)
+        new_pos_embed = torch.cat((extra_tokens, pos_tokens), dim=1)
+        print('reshape position embedding from %d to %d'%(orig_size ** 2,new_size ** 2))
+        
+        return new_pos_embed    
+    else:
+        return pos_embed_checkpoint

extras/GroundingDINO/config/GroundingDINO_SwinT_OGC.py

@@ -0,0 +1,43 @@
+batch_size = 1
+modelname = "groundingdino"
+backbone = "swin_T_224_1k"
+position_embedding = "sine"
+pe_temperatureH = 20
+pe_temperatureW = 20
+return_interm_indices = [1, 2, 3]
+backbone_freeze_keywords = None
+enc_layers = 6
+dec_layers = 6
+pre_norm = False
+dim_feedforward = 2048
+hidden_dim = 256
+dropout = 0.0
+nheads = 8
+num_queries = 900
+query_dim = 4
+num_patterns = 0
+num_feature_levels = 4
+enc_n_points = 4
+dec_n_points = 4
+two_stage_type = "standard"
+two_stage_bbox_embed_share = False
+two_stage_class_embed_share = False
+transformer_activation = "relu"
+dec_pred_bbox_embed_share = True
+dn_box_noise_scale = 1.0
+dn_label_noise_ratio = 0.5
+dn_label_coef = 1.0
+dn_bbox_coef = 1.0
+embed_init_tgt = True
+dn_labelbook_size = 2000
+max_text_len = 256
+text_encoder_type = "bert-base-uncased"
+use_text_enhancer = True
+use_fusion_layer = True
+use_checkpoint = True
+use_transformer_ckpt = True
+use_text_cross_attention = True
+text_dropout = 0.0
+fusion_dropout = 0.0
+fusion_droppath = 0.1
+sub_sentence_present = True

extras/GroundingDINO/util/inference.py

@@ -0,0 +1,100 @@
+from typing import Tuple, List
+
+import ldm_patched.modules.model_management as model_management
+from ldm_patched.modules.model_patcher import ModelPatcher
+from modules.config import path_inpaint
+from modules.model_loader import load_file_from_url
+
+import numpy as np
+import supervision as sv
+import torch
+from groundingdino.util.inference import Model
+from groundingdino.util.inference import load_model, preprocess_caption, get_phrases_from_posmap
+
+
+class GroundingDinoModel(Model):
+    def __init__(self):
+        self.config_file = 'extras/GroundingDINO/config/GroundingDINO_SwinT_OGC.py'
+        self.model = None
+        self.load_device = torch.device('cpu')
+        self.offload_device = torch.device('cpu')
+
+    @torch.no_grad()
+    @torch.inference_mode()
+    def predict_with_caption(
+            self,
+            image: np.ndarray,
+            caption: str,
+            box_threshold: float = 0.35,
+            text_threshold: float = 0.25
+    ) -> Tuple[sv.Detections, torch.Tensor, torch.Tensor, List[str]]:
+        if self.model is None:
+            filename = load_file_from_url(
+                url="https://github.com/IDEA-Research/GroundingDINO/releases/download/v0.1.0-alpha/groundingdino_swint_ogc.pth",
+                file_name='groundingdino_swint_ogc.pth',
+                model_dir=path_inpaint)
+            model = load_model(model_config_path=self.config_file, model_checkpoint_path=filename)
+
+            self.load_device = model_management.text_encoder_device()
+            self.offload_device = model_management.text_encoder_offload_device()
+
+            model.to(self.offload_device)
+
+            self.model = ModelPatcher(model, load_device=self.load_device, offload_device=self.offload_device)
+
+        model_management.load_model_gpu(self.model)
+
+        processed_image = GroundingDinoModel.preprocess_image(image_bgr=image).to(self.load_device)
+        boxes, logits, phrases = predict(
+            model=self.model,
+            image=processed_image,
+            caption=caption,
+            box_threshold=box_threshold,
+            text_threshold=text_threshold,
+            device=self.load_device)
+        source_h, source_w, _ = image.shape
+        detections = GroundingDinoModel.post_process_result(
+            source_h=source_h,
+            source_w=source_w,
+            boxes=boxes,
+            logits=logits)
+        return detections, boxes, logits, phrases
+
+
+def predict(
+        model,
+        image: torch.Tensor,
+        caption: str,
+        box_threshold: float,
+        text_threshold: float,
+        device: str = "cuda"
+) -> Tuple[torch.Tensor, torch.Tensor, List[str]]:
+    caption = preprocess_caption(caption=caption)
+
+    # override to use model wrapped by patcher
+    model = model.model.to(device)
+    image = image.to(device)
+
+    with torch.no_grad():
+        outputs = model(image[None], captions=[caption])
+
+    prediction_logits = outputs["pred_logits"].cpu().sigmoid()[0]  # prediction_logits.shape = (nq, 256)
+    prediction_boxes = outputs["pred_boxes"].cpu()[0]  # prediction_boxes.shape = (nq, 4)
+
+    mask = prediction_logits.max(dim=1)[0] > box_threshold
+    logits = prediction_logits[mask]  # logits.shape = (n, 256)
+    boxes = prediction_boxes[mask]  # boxes.shape = (n, 4)
+
+    tokenizer = model.tokenizer
+    tokenized = tokenizer(caption)
+
+    phrases = [
+        get_phrases_from_posmap(logit > text_threshold, tokenized, tokenizer).replace('.', '')
+        for logit
+        in logits
+    ]
+
+    return boxes, logits.max(dim=1)[0], phrases
+
+
+default_groundingdino = GroundingDinoModel().predict_with_caption

extras/censor.py

@@ -0,0 +1,60 @@
+import os
+
+import numpy as np
+import torch
+from transformers import CLIPConfig, CLIPImageProcessor
+
+import ldm_patched.modules.model_management as model_management
+import modules.config
+from extras.safety_checker.models.safety_checker import StableDiffusionSafetyChecker
+from ldm_patched.modules.model_patcher import ModelPatcher
+
+safety_checker_repo_root = os.path.join(os.path.dirname(__file__), 'safety_checker')
+config_path = os.path.join(safety_checker_repo_root, "configs", "config.json")
+preprocessor_config_path = os.path.join(safety_checker_repo_root, "configs", "preprocessor_config.json")
+
+
+class Censor:
+    def __init__(self):
+        self.safety_checker_model: ModelPatcher | None = None
+        self.clip_image_processor: CLIPImageProcessor | None = None
+        self.load_device = torch.device('cpu')
+        self.offload_device = torch.device('cpu')
+
+    def init(self):
+        if self.safety_checker_model is None and self.clip_image_processor is None:
+            safety_checker_model = modules.config.downloading_safety_checker_model()
+            self.clip_image_processor = CLIPImageProcessor.from_json_file(preprocessor_config_path)
+            clip_config = CLIPConfig.from_json_file(config_path)
+            model = StableDiffusionSafetyChecker.from_pretrained(safety_checker_model, config=clip_config)
+            model.eval()
+
+            self.load_device = model_management.text_encoder_device()
+            self.offload_device = model_management.text_encoder_offload_device()
+
+            model.to(self.offload_device)
+
+            self.safety_checker_model = ModelPatcher(model, load_device=self.load_device, offload_device=self.offload_device)
+
+    def censor(self, images: list | np.ndarray) -> list | np.ndarray:
+        self.init()
+        model_management.load_model_gpu(self.safety_checker_model)
+
+        single = False
+        if not isinstance(images, (list, np.ndarray)):
+            images = [images]
+            single = True
+
+        safety_checker_input = self.clip_image_processor(images, return_tensors="pt")
+        safety_checker_input.to(device=self.load_device)
+        checked_images, has_nsfw_concept = self.safety_checker_model.model(images=images,
+                                                                           clip_input=safety_checker_input.pixel_values)
+        checked_images = [image.astype(np.uint8) for image in checked_images]
+
+        if single:
+            checked_images = checked_images[0]
+
+        return checked_images
+
+
+default_censor = Censor().censor

extras/expansion.py

@@ -0,0 +1,129 @@
+# Fooocus GPT2 Expansion
+# Algorithm created by Lvmin Zhang at 2023, Stanford
+# If used inside Fooocus, any use is permitted.
+# If used outside Fooocus, only non-commercial use is permitted (CC-By NC 4.0).
+# This applies to the word list, vocab, model, and algorithm.
+
+
+import os
+import torch
+import math
+import ldm_patched.modules.model_management as model_management
+
+from transformers.generation.logits_process import LogitsProcessorList
+from transformers import AutoTokenizer, AutoModelForCausalLM, set_seed
+from modules.config import path_fooocus_expansion
+from ldm_patched.modules.model_patcher import ModelPatcher
+
+
+# limitation of np.random.seed(), called from transformers.set_seed()
+SEED_LIMIT_NUMPY = 2**32
+neg_inf = - 8192.0
+
+
+def safe_str(x):
+    x = str(x)
+    for _ in range(16):
+        x = x.replace('  ', ' ')
+    return x.strip(",. \r\n")
+
+
+def remove_pattern(x, pattern):
+    for p in pattern:
+        x = x.replace(p, '')
+    return x
+
+
+class FooocusExpansion:
+    def __init__(self):
+        self.tokenizer = AutoTokenizer.from_pretrained(path_fooocus_expansion)
+
+        positive_words = open(os.path.join(path_fooocus_expansion, 'positive.txt'),
+                              encoding='utf-8').read().splitlines()
+        positive_words = ['Ġ' + x.lower() for x in positive_words if x != '']
+
+        self.logits_bias = torch.zeros((1, len(self.tokenizer.vocab)), dtype=torch.float32) + neg_inf
+
+        debug_list = []
+        for k, v in self.tokenizer.vocab.items():
+            if k in positive_words:
+                self.logits_bias[0, v] = 0
+                debug_list.append(k[1:])
+
+        print(f'Fooocus V2 Expansion: Vocab with {len(debug_list)} words.')
+
+        # debug_list = '\n'.join(sorted(debug_list))
+        # print(debug_list)
+
+        # t11 = self.tokenizer(',', return_tensors="np")
+        # t198 = self.tokenizer('\n', return_tensors="np")
+        # eos = self.tokenizer.eos_token_id
+
+        self.model = AutoModelForCausalLM.from_pretrained(path_fooocus_expansion)
+        self.model.eval()
+
+        load_device = model_management.text_encoder_device()
+        offload_device = model_management.text_encoder_offload_device()
+
+        # MPS hack
+        if model_management.is_device_mps(load_device):
+            load_device = torch.device('cpu')
+            offload_device = torch.device('cpu')
+
+        use_fp16 = model_management.should_use_fp16(device=load_device)
+
+        if use_fp16:
+            self.model.half()
+
+        self.patcher = ModelPatcher(self.model, load_device=load_device, offload_device=offload_device)
+        print(f'Fooocus Expansion engine loaded for {load_device}, use_fp16 = {use_fp16}.')
+
+    @torch.no_grad()
+    @torch.inference_mode()
+    def logits_processor(self, input_ids, scores):
+        assert scores.ndim == 2 and scores.shape[0] == 1
+        self.logits_bias = self.logits_bias.to(scores)
+
+        bias = self.logits_bias.clone()
+        bias[0, input_ids[0].to(bias.device).long()] = neg_inf
+        bias[0, 11] = 0
+
+        return scores + bias
+
+    @torch.no_grad()
+    @torch.inference_mode()
+    def __call__(self, prompt, seed):
+        if prompt == '':
+            return ''
+
+        if self.patcher.current_device != self.patcher.load_device:
+            print('Fooocus Expansion loaded by itself.')
+            model_management.load_model_gpu(self.patcher)
+
+        seed = int(seed) % SEED_LIMIT_NUMPY
+        set_seed(seed)
+        prompt = safe_str(prompt) + ','
+
+        tokenized_kwargs = self.tokenizer(prompt, return_tensors="pt")
+        tokenized_kwargs.data['input_ids'] = tokenized_kwargs.data['input_ids'].to(self.patcher.load_device)
+        tokenized_kwargs.data['attention_mask'] = tokenized_kwargs.data['attention_mask'].to(self.patcher.load_device)
+
+        current_token_length = int(tokenized_kwargs.data['input_ids'].shape[1])
+        max_token_length = 75 * int(math.ceil(float(current_token_length) / 75.0))
+        max_new_tokens = max_token_length - current_token_length
+
+        if max_new_tokens == 0:
+            return prompt[:-1]
+
+        # https://huggingface.co/blog/introducing-csearch
+        # https://huggingface.co/docs/transformers/generation_strategies
+        features = self.model.generate(**tokenized_kwargs,
+                                       top_k=100,
+                                       max_new_tokens=max_new_tokens,
+                                       do_sample=True,
+                                       logits_processor=LogitsProcessorList([self.logits_processor]))
+
+        response = self.tokenizer.batch_decode(features, skip_special_tokens=True)
+        result = safe_str(response[0])
+
+        return result

extras/face_crop.py

@@ -0,0 +1,50 @@
+import cv2
+import numpy as np
+import modules.config
+
+
+faceRestoreHelper = None
+
+
+def align_warp_face(self, landmark, border_mode='constant'):
+    affine_matrix = cv2.estimateAffinePartial2D(landmark, self.face_template, method=cv2.LMEDS)[0]
+    self.affine_matrices.append(affine_matrix)
+    if border_mode == 'constant':
+        border_mode = cv2.BORDER_CONSTANT
+    elif border_mode == 'reflect101':
+        border_mode = cv2.BORDER_REFLECT101
+    elif border_mode == 'reflect':
+        border_mode = cv2.BORDER_REFLECT
+    input_img = self.input_img
+    cropped_face = cv2.warpAffine(input_img, affine_matrix, self.face_size,
+                                  borderMode=border_mode, borderValue=(135, 133, 132))
+    return cropped_face
+
+
+def crop_image(img_rgb):
+    global faceRestoreHelper
+    
+    if faceRestoreHelper is None:
+        from extras.facexlib.utils.face_restoration_helper import FaceRestoreHelper
+        faceRestoreHelper = FaceRestoreHelper(
+            upscale_factor=1,
+            model_rootpath=modules.config.path_controlnet,
+            device='cpu'  # use cpu is safer since we are out of memory management
+        )
+
+    faceRestoreHelper.clean_all()
+    faceRestoreHelper.read_image(np.ascontiguousarray(img_rgb[:, :, ::-1].copy()))
+    faceRestoreHelper.get_face_landmarks_5()
+
+    landmarks = faceRestoreHelper.all_landmarks_5
+    # landmarks are already sorted with confidence.
+
+    if len(landmarks) == 0:
+        print('No face detected')
+        return img_rgb
+    else:
+        print(f'Detected {len(landmarks)} faces')
+
+    result = align_warp_face(faceRestoreHelper, landmarks[0])
+
+    return np.ascontiguousarray(result[:, :, ::-1].copy())