Use larger model as default model for non-space and non-openai solution

README.md

@@ -27,7 +27,7 @@ Check out the configuration reference at https://huggingface.co/docs/hub/spaces-
 python app.py
 # Streamlit (TODO: able to clear message history)
 streamlit run streamlit_app.py
-# Streamlit chat_input (TODO: move ReAct process to `st.expander`)
+# Streamlit chat_input (TODO: move ReAct process to `st.expander`; able to render image)
 streamlit run streamlit_app_chat.py
 # Chainlit (TODO: move the ReAct process from agent message to "step")
 chainlit run chainlit_app.py
@@ -38,6 +38,9 @@ chainlit run chainlit_app.py
 - `What time is it?`
   - Expect to use tool `get_current_time_in_timezone`
   - Result expect the current local time in UTC
+- `Draw a bear for me`
+  - Expect to use tool `image_generator`
+  - Result will be something like `{'path': '/var/folders/g3/gsbp4bsx1gs51t1805r34ztw0000gn/T/tmpch4mrr13/2fa94447-e9e5-4964-b9d2-b4b9a83ccf84.png', 'mime_type': 'image/png'}`
 
 ---
 

app.py

@@ -62,7 +62,7 @@ if IS_IN_HF_SPACE := os.getenv("SPACE_ID"):
     print("Using HfApiModel with model_id:", model_id)
 
     model = HfApiModel(
-        max_tokens=2096,
+        max_tokens=2096 if IS_IN_HF_SPACE else None,
         temperature=0.5,
         model_id=model_id,
         custom_role_conversions=None,
@@ -80,16 +80,29 @@ else:
 
         model = OpenAIServerModel(model_id="gpt-3.5-turbo", api_key=OPENAI_API_KEY)
     else:
-        # NOTE: this model is not good enough for agent
+        # NOTE: this model is not good enough for agent (and still might be too heavy to run on a normal computer)
+        # print(
+        #     "Using TransformersModel with model_id: HuggingFaceTB/SmolLM2-1.7B-Instruct"
+        # )
+        #
+        # model = TransformersModel(
+        #     model_id="HuggingFaceTB/SmolLM2-1.7B-Instruct", trust_remote_code=True
+        # )
+
+        # NOTE: this model is well enough to use simple tools (but need authentication)
+        # https://huggingface.co/meta-llama/Llama-3.2-11B-Vision-Instruct
+        # https://huggingface.co/settings/gated-repos
         print(
-            "Using TransformersModel with model_id: HuggingFaceTB/SmolLM2-1.7B-Instruct"
+            "Using HfApiModel with model_id: meta-llama/Llama-3.2-11B-Vision-Instruct"
         )
 
-        model = TransformersModel(
-            model_id="HuggingFaceTB/SmolLM2-1.7B-Instruct", trust_remote_code=True
+        # NOTE: this use InferenceClient under the hood
+        model = HfApiModel(
+            temperature=0.5,
+            model_id="meta-llama/Llama-3.2-11B-Vision-Instruct",
+            custom_role_conversions=None,
         )
 
-
 # Import tool from Hub
 image_generation_tool = load_tool("agents-course/text-to-image", trust_remote_code=True)