Move to in-library checkpoint (#60)

- Move to in-library checkpoint (bb9e6c53c434569bce78be43abf05d629735328f)
- Preparations for transition to in-library checkpoint (972d3ef44362a0c96e89fd210e570ab0741b34b7)
- Fix typo (50ec42273167020bc69d4aa0f7e5657b5b72fe5b)
- Revert to Falcon naming (eddf7c444f833df31722d4745ff46e0b3f502723)

README.md

@@ -21,6 +21,8 @@ license: apache-2.0
 * **Falcon-40B is the best open-source model available.** It outperforms [LLaMA](https://github.com/facebookresearch/llama), [StableLM](https://github.com/Stability-AI/StableLM), [RedPajama](https://huggingface.co/togethercomputer/RedPajama-INCITE-Base-7B-v0.1), [MPT](https://huggingface.co/mosaicml/mpt-7b), etc. See the [OpenLLM Leaderboard](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard).
 * **It features an architecture optimized for inference**, with FlashAttention ([Dao et al., 2022](https://arxiv.org/abs/2205.14135)) and multiquery ([Shazeer et al., 2019](https://arxiv.org/abs/1911.02150)). 
 
+⚠️ Falcon is now available as a core model in the `transformers` library! To use the in-library version, please install the latest version of `transformers` with `pip install git+https://github.com/huggingface/transformers.git`, then simply remove the `trust_remote_code=True` argument from `from_pretrained()`.
+
 💬 **This is an instruct model, which may not be ideal for further finetuning.** If you are interested in building your own instruct/chat model, we recommend starting from [Falcon-40B](https://huggingface.co/tiiuae/falcon-40b). 
 
 💸 **Looking for a smaller, less expensive model?** [Falcon-7B-Instruct](https://huggingface.co/tiiuae/falcon-7b-instruct) is Falcon-40B-Instruct's little brother!
@@ -38,7 +40,6 @@ pipeline = transformers.pipeline(
     model=model,
     tokenizer=tokenizer,
     torch_dtype=torch.bfloat16,
-    trust_remote_code=True,
     device_map="auto",
 )
 sequences = pipeline(
@@ -110,7 +111,6 @@ pipeline = transformers.pipeline(
     model=model,
     tokenizer=tokenizer,
     torch_dtype=torch.bfloat16,
-    trust_remote_code=True,
     device_map="auto",
 )
 sequences = pipeline(
@@ -219,4 +219,4 @@ To cite the [Baize](https://github.com/project-baize/baize-chatbot) instruction
 Falcon-40B-Instruct is made available under the Apache 2.0 license.
 
 ## Contact
-[email protected]
+[email protected]

config.json

@@ -2,12 +2,16 @@
   "alibi": false,
   "apply_residual_connection_post_layernorm": false,
   "architectures": [
-    "RWForCausalLM"
+    "FalconForCausalLM"
   ],
   "attention_dropout": 0.0,
   "auto_map": {
-    "AutoConfig": "configuration_RW.RWConfig",
-    "AutoModelForCausalLM": "modelling_RW.RWForCausalLM"
+    "AutoConfig": "configuration_falcon.FalconConfig",
+    "AutoModel": "modeling_falcon.FalconModel",
+    "AutoModelForSequenceClassification": "modeling_falcon.FalconForSequenceClassification",
+    "AutoModelForTokenClassification": "modeling_falcon.FalconForTokenClassification",
+    "AutoModelForQuestionAnswering": "modeling_falcon.FalconForQuestionAnswering",
+    "AutoModelForCausalLM": "modeling_falcon.FalconForCausalLM"
   },
   "bias": false,
   "bos_token_id": 11,
@@ -16,10 +20,11 @@
   "hidden_size": 8192,
   "initializer_range": 0.02,
   "layer_norm_epsilon": 1e-05,
-  "model_type": "RefinedWeb",
-  "n_head": 128,
-  "n_head_kv": 8,
-  "n_layer": 60,
+  "model_type": "falcon",
+  "new_decoder_architecture": true,
+  "num_attention_heads": 128,
+  "num_hidden_layers": 60,
+  "num_kv_heads": 8,
   "parallel_attn": true,
   "torch_dtype": "bfloat16",
   "transformers_version": "4.26.0",

configuration_RW.py

@@ -1,75 +0,0 @@
-# coding=utf-8
-# Copyright 2022 the Big Science Workshop and HuggingFace Inc. team.  All rights reserved.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-""" Bloom configuration"""
-from transformers.configuration_utils import PretrainedConfig
-from transformers.utils import logging
-
-
-logger = logging.get_logger(__name__)
-
-
-class RWConfig(PretrainedConfig):
-    model_type = "RefinedWeb"
-    keys_to_ignore_at_inference = ["past_key_values"]
-    attribute_map = {
-        "num_hidden_layers": "n_layer",
-        "num_attention_heads": "n_head",
-    }
-
-    def __init__(
-        self,
-        vocab_size=250880,
-        hidden_size=64,
-        n_layer=2,
-        n_head=8,
-        layer_norm_epsilon=1e-5,
-        initializer_range=0.02,
-        use_cache=True,
-        bos_token_id=1,
-        eos_token_id=2,
-        apply_residual_connection_post_layernorm=False,
-        hidden_dropout=0.0,
-        attention_dropout=0.0,
-        n_head_kv=None,
-        alibi=False,
-        **kwargs,
-    ):
-        self.vocab_size = vocab_size
-        # Backward compatibility with n_embed kwarg
-        n_embed = kwargs.pop("n_embed", None)
-        self.hidden_size = hidden_size if n_embed is None else n_embed
-        self.n_layer = n_layer
-        self.n_head = n_head
-        self.layer_norm_epsilon = layer_norm_epsilon
-        self.initializer_range = initializer_range
-        self.use_cache = use_cache
-        self.apply_residual_connection_post_layernorm = apply_residual_connection_post_layernorm
-        self.hidden_dropout = hidden_dropout
-        self.attention_dropout = attention_dropout
-
-        self.bos_token_id = bos_token_id
-        self.eos_token_id = eos_token_id
-        self.n_head_kv = n_head if n_head_kv is None else n_head_kv
-        self.alibi = alibi
-
-        super().__init__(bos_token_id=bos_token_id, eos_token_id=eos_token_id, **kwargs)
-
-    @property
-    def head_dim(self):
-        return self.hidden_size // self.n_head
-
-    @property
-    def rotary(self):
-        return not self.alibi

configuration_falcon.py

@@ -0,0 +1,147 @@
+# coding=utf-8
+# Copyright 2023 the Falcon authors and HuggingFace Inc. team.  All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+""" Falcon configuration"""
+from transformers.configuration_utils import PretrainedConfig
+from transformers.utils import logging
+
+
+logger = logging.get_logger(__name__)
+
+FALCON_PRETRAINED_CONFIG_ARCHIVE_MAP = {
+    "tiiuae/falcon-40b": "https://huggingface.co/tiiuae/falcon-40b/resolve/main/config.json",
+    "tiiuae/falcon-7b": "https://huggingface.co/tiiuae/falcon-7b/resolve/main/config.json",
+}
+
+
+class FalconConfig(PretrainedConfig):
+    r"""
+    This is the configuration class to store the configuration of a [`FalconModel`]. It is used to instantiate a Falcon
+    model according to the specified arguments, defining the model architecture. Instantiating a configuration with the
+    defaults will yield a similar configuration to that of the
+    [tiiuae/falcon-7b](https://huggingface.co/tiiuae/falcon-7b) architecture.
+
+    Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
+    documentation from [`PretrainedConfig`] for more information.
+
+
+    Args:
+        vocab_size (`int`, *optional*, defaults to 65024):
+            Vocabulary size of the Falcon model. Defines the number of different tokens that can be represented by the
+            `inputs_ids` passed when calling [`FalconModel`]
+        hidden_size (`int`, *optional*, defaults to 4544):
+            Dimension of the hidden representations.
+        num_hidden_layers (`int`, *optional*, defaults to 32):
+            Number of hidden layers in the Transformer decoder.
+        num_attention_heads (`int`, *optional*, defaults to 71):
+            Number of attention heads for each attention layer in the Transformer encoder.
+        initializer_range (`float`, *optional*, defaults to 0.02):
+            The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
+        use_cache (`bool`, *optional*, defaults to `True`):
+            Whether the model should return the last key/values attentions (not used by all models). Only relevant if
+            `config.is_decoder=True`.
+        layer_norm_epsilon (`float`, *optional*, defaults to 1e-5):
+            The epsilon used by the layer normalization layers.
+        hidden_dropout (`float`, *optional*, defaults to 0.0):
+            The dropout probability for MLP layers.
+        attention_dropout (`float`, *optional*, defaults to 0.0):
+            The dropout probability for attention layers.
+        num_kv_heads (`int`, *optional*):
+            Number of key-value heads to use per attention layer. If unset, defaults to the same value as
+            `num_attention_heads`.
+        alibi (`bool`, *optional*, defaults to `False`):
+            Whether to use ALiBi positional biases during self-attention.
+        new_decoder_architecture (`bool`, *optional*, defaults to `False`):
+            Whether to use the new (Falcon-40B) decoder architecture. If `True`, the `multi_query` and `parallel_attn`
+            arguments are ignored, as the new decoder always uses parallel attention.
+        multi_query (`bool`, *optional*, defaults to `True`):
+            Whether to use multi-query attention in the decoder. Ignored when `new_decoder_architecture` is `True`.
+        parallel_attn (`bool`, *optional*, defaults to `True`):
+            Whether to compute attention in parallel with the feedforward layer. If False, they are consecutive
+            instead, as in the original Transformer architecture. Ignored when `new_decoder_architecture` is `True`.
+        bias (`bool`, *optional*, defaults to `False`):
+            Whether to use bias on Linear layers.
+        bos_token_id (`int`, *optional*, defaults to 11):
+            The id of the "beginning-of-sequence" token.
+        eos_token_id (`int`, *optional*, defaults to 11):
+            The id of the "end-of-sequence" token.
+
+    Example:
+
+    ```python
+    >>> from transformers import FalconModel, FalconConfig
+
+    >>> # Initializing a small (2-layer) Falcon configuration
+    >>> configuration = FalconConfig(num_hidden_layers=2)
+
+    >>> # Initializing a model from the small configuration
+    >>> model = FalconModel(configuration)
+
+    >>> # Accessing the model configuration
+    >>> configuration = model.config
+    ```"""
+    model_type = "falcon"
+    keys_to_ignore_at_inference = ["past_key_values"]
+
+    def __init__(
+        self,
+        vocab_size=65024,
+        hidden_size=4544,
+        num_hidden_layers=32,
+        num_attention_heads=71,
+        layer_norm_epsilon=1e-5,
+        initializer_range=0.02,
+        use_cache=True,
+        hidden_dropout=0.0,
+        attention_dropout=0.0,
+        num_kv_heads=None,
+        alibi=False,
+        new_decoder_architecture=False,
+        multi_query=True,
+        parallel_attn=True,
+        bias=False,
+        bos_token_id=11,
+        eos_token_id=11,
+        **kwargs,
+    ):
+        self.vocab_size = vocab_size
+        # Backward compatibility with n_embed kwarg
+        n_embed = kwargs.pop("n_embed", None)
+        self.hidden_size = hidden_size if n_embed is None else n_embed
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+        self.layer_norm_epsilon = layer_norm_epsilon
+        self.initializer_range = initializer_range
+        self.use_cache = use_cache
+        self.hidden_dropout = hidden_dropout
+        self.attention_dropout = attention_dropout
+
+        self.bos_token_id = bos_token_id
+        self.eos_token_id = eos_token_id
+        self.num_kv_heads = num_attention_heads if num_kv_heads is None else num_kv_heads
+        self.alibi = alibi
+        self.new_decoder_architecture = new_decoder_architecture
+        self.multi_query = multi_query  # Ignored when new_decoder_architecture is True
+        self.parallel_attn = parallel_attn
+        self.bias = bias
+
+        super().__init__(bos_token_id=bos_token_id, eos_token_id=eos_token_id, **kwargs)
+
+    @property
+    def head_dim(self):
+        return self.hidden_size // self.num_attention_heads
+
+    @property
+    def rotary(self):
+        return not self.alibi

generation_config.json

@@ -1,6 +1,6 @@
 {
   "_from_model_config": true,
-  "bos_token_id": 1,
-  "eos_token_id": 2,
-  "transformers_version": "4.26.0"
-}
+  "bos_token_id": 11,
+  "eos_token_id": 11,
+  "transformers_version": "4.31.0.dev0"
+}

modelling_RW.py → modeling_falcon.py

@@ -1,9 +1,20 @@
-# port of models described in RW
-# We use the bloom model as a starting point for these model.
-# Please refer to the bloom models for usage instructions.
+# coding=utf-8
+# Copyright 2023 the Falcon authors and HuggingFace Inc. team.  All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""PyTorch Falcon model."""
 
 import math
-import warnings
 from typing import Optional, Tuple, Union
 
 import torch
@@ -20,59 +31,60 @@ from transformers.modeling_outputs import (
     TokenClassifierOutput,
 )
 from transformers.modeling_utils import PreTrainedModel
-from transformers.utils import logging
-from .configuration_RW import RWConfig
+from transformers.utils import add_code_sample_docstrings, add_start_docstrings, add_start_docstrings_to_model_forward, logging
+from .configuration_falcon import FalconConfig
+
 
 logger = logging.get_logger(__name__)
 
+FALCON_PRETRAINED_MODEL_ARCHIVE_LIST = [
+    "tiiuae/falcon-40b",
+    "tiiuae/falcon-40b-instruct",
+    "tiiuae/falcon-7b",
+    "tiiuae/falcon-7b-instruct",
+    "tiiuae/falcon-rw-7b",
+    "tiiuae/falcon-rw-1b",
+]
+_CHECKPOINT_FOR_DOC = "Rocketknight1/falcon-rw-1b"
+_CONFIG_FOR_DOC = "FalconConfig"
+
+
 # NOTE(Hesslow): Unfortunately we did not fuse matmul and bias during training, this means that there's one additional quantization to bfloat16 between the operations.
 # In order not to degrade the quality of our HF-port, we keep these characteristics in the final model.
-class Linear(nn.Linear):
+class FalconLinear(nn.Linear):
     def forward(self, input: torch.Tensor) -> torch.Tensor:
-        ret = input @ self.weight.T
+        hidden_states = input @ self.weight.T
         if self.bias is None:
-            return ret
-        else:
-            return ret + self.bias
-
+            return hidden_states
+        return hidden_states + self.bias
 
-from einops import rearrange
 
 # rotary pos emb helpers (torch.jit.script does not seem to support staticmethod...)
 def rotate_half(x):
     x1, x2 = x[..., : x.shape[-1] // 2], x[..., x.shape[-1] // 2 :]
-    return torch.cat((-x2, x1), dim=x1.ndim - 1)  # dim=-1 triggers a bug in torch < 1.8.0
+    return torch.cat((-x2, x1), dim=-1)
 
 
-class RotaryEmbedding(torch.nn.Module):
+class FalconRotaryEmbedding(nn.Module):
     """Implementation of RotaryEmbedding from GPT-NeoX.
-    This implementation is design to operate on queries and keys that are compatible with
-    [batch_size, n_heads_per_partition, seq_len, head_dim] (e.g. MinGPTAttention format).
+    This implementation is designed to operate on queries and keys that are compatible with `[batch_size,
+    n_heads_per_partition, seq_len, head_dim]` (e.g. MinGPTAttention format).
     """
 
-    def __init__(
-        self,
-        head_dim: int,
-        base=10000,
-    ):
+    def __init__(self, head_dim: int, base=10000):
         super().__init__()
         inv_freq = 1.0 / (base ** (torch.arange(0, head_dim, 2).float() / head_dim))
         self.register_buffer("inv_freq", inv_freq, persistent=False)
         self.head_dim = head_dim
-        self.seq_len_cached = None
-        self.batch_size_cached = None
+        self.seq_len_cached = -1
         self.cos_cached: torch.Tensor | None = None
         self.sin_cached: torch.Tensor | None = None
 
-    def cos_sin(
-        self,
-        seq_len: int,
-        device="cuda",
-        dtype=torch.bfloat16,
-    ) -> torch.Tensor:
-        if seq_len != self.seq_len_cached:
-            self.seq_len_cached = seq_len
-            t = torch.arange(seq_len, device=device).type_as(self.inv_freq)
+    def cos_sin(self, seq_len: int, past_key_values_length: int, device="cpu", dtype=torch.bfloat16) -> torch.Tensor:
+        total_length = seq_len + past_key_values_length
+        if total_length > self.seq_len_cached:
+            self.seq_len_cached = total_length
+            t = torch.arange(total_length, device=device, dtype=self.inv_freq.dtype)
             freqs = torch.einsum("i,j->ij", t, self.inv_freq)
             emb = torch.cat((freqs, freqs), dim=-1).to(device)
 
@@ -85,36 +97,46 @@ class RotaryEmbedding(torch.nn.Module):
             self.cos_cached = self.cos_cached.type(dtype)
             self.sin_cached = self.sin_cached.type(dtype)
 
-        return self.cos_cached, self.sin_cached
+        return (
+            self.cos_cached[:, past_key_values_length : seq_len + past_key_values_length],
+            self.sin_cached[:, past_key_values_length : seq_len + past_key_values_length],
+        )
 
-    def forward(self, q, k):
-        batch, seq_len, head_dim = q.shape
-        cos, sin = self.cos_sin(seq_len, q.device, q.dtype)
-        return (q * cos) + (rotate_half(q) * sin), (k * cos) + (rotate_half(k) * sin)
+    def forward(self, query, key, past_key_values_length=0):
+        batch, seq_len, head_dim = query.shape
+        cos, sin = self.cos_sin(seq_len, past_key_values_length, query.device, query.dtype)
+        return (query * cos) + (rotate_half(query) * sin), (key * cos) + (rotate_half(key) * sin)
 
 
 def _make_causal_mask(
     input_ids_shape: torch.Size, device: torch.device, past_key_values_length: int
 ) -> torch.BoolTensor:
+    """
+    Make causal mask used for self-attention. This mask does not take the existing attention mask into account - it
+    just blocks tokens from attending forwards in the sequence. The output shape will be `[batch_size, 1,
+    target_length, target_length+past_key_values_length]`.
+    """
     batch_size, target_length = input_ids_shape
-    mask = torch.empty((target_length, target_length + past_key_values_length), dtype=torch.bool, device=device)
-    # ONNX doesn't support `torch.Tensor.triu` properly, thus we use this workaround
-    seq_ids = torch.arange(target_length, device=device)
-    mask[:, past_key_values_length:] = seq_ids[:, None] < seq_ids[None, :]
-
-    if past_key_values_length > 0:
-        mask[:, :past_key_values_length] = False
 
+    mask = torch.triu(torch.ones((target_length, target_length), dtype=torch.bool, device=device), diagonal=1)
+    # If past_key_values_length is 0 this is an empty tensor and the concatenation is a no-op.
+    # This code style is an unfortunate consequence of getting your TF engineer to port models; doing it this
+    # way avoids a data-dependent conditional, which will help me when I have to port this to XLA later.
+    past_mask = torch.zeros((target_length, past_key_values_length), dtype=torch.bool, device=device)
+    mask = torch.cat([past_mask, mask], dim=-1)
     expanded_mask = mask[None, None, :, :].expand(batch_size, 1, target_length, target_length + past_key_values_length)
     return expanded_mask
 
 
-def _expand_mask(mask: torch.Tensor, tgt_length: int) -> torch.BoolTensor:
-    batch_size, src_length = mask.shape
-    tgt_length = tgt_length if tgt_length is not None else src_length
+def _expand_mask(mask: torch.Tensor, past_key_values_length: int) -> torch.BoolTensor:
+    """
+    Expands attention_mask from `[batch_size, seq_length]` to `[batch_size, 1, seq_length, seq_length + past_length]`.
+    """
+    batch_size, total_length = mask.shape
+    seq_length = total_length - past_key_values_length if past_key_values_length is not None else total_length
 
     expanded_mask = ~(mask[:, None, None, :].to(torch.bool))
-    return expanded_mask.expand(batch_size, 1, tgt_length, src_length)
+    return expanded_mask.expand(batch_size, 1, seq_length, total_length)
 
 
 def build_alibi_tensor(attention_mask: torch.Tensor, num_heads: int, dtype: torch.dtype) -> torch.Tensor:
@@ -145,18 +167,32 @@ def build_alibi_tensor(attention_mask: torch.Tensor, num_heads: int, dtype: torc
     return alibi.reshape(batch_size * num_heads, 1, seq_length).to(dtype)
 
 
+# Copied from transformers.models.bloom.modeling_bloom.dropout_add
 def dropout_add(x: torch.Tensor, residual: torch.Tensor, prob: float, training: bool) -> torch.Tensor:
+    """
+    Dropout add function
+
+    Args:
+        x (`torch.tensor`, *required*):
+            input tensor
+        residual (`torch.tensor`, *required*):
+            residual tensor
+        prob (`float`, *required*):
+            dropout probability
+        training (`bool`, *required*):
+            training mode
+    """
     out = F.dropout(x, p=prob, training=training)
     out = residual + out
     return out
 
 
-class Attention(nn.Module):
-    def __init__(self, config: RWConfig):
+class FalconAttention(nn.Module):
+    def __init__(self, config: FalconConfig):
         super().__init__()
 
         self.hidden_size = config.hidden_size
-        self.num_heads = config.n_head
+        self.num_heads = config.num_attention_heads
         self.head_dim = self.hidden_size // self.num_heads
         self.split_size = self.hidden_size
         self.hidden_dropout = config.hidden_dropout
@@ -167,59 +203,62 @@ class Attention(nn.Module):
                 f" {self.num_heads})."
             )
 
-        self.maybe_rotary = RotaryEmbedding(config.head_dim) if config.rotary else lambda q, k: (q, k)
+        self.maybe_rotary = FalconRotaryEmbedding(config.head_dim) if config.rotary else lambda q, k, t: (q, k)
 
         # Layer-wise attention scaling
         self.inv_norm_factor = 1.0 / math.sqrt(self.head_dim)
         self.beta = self.inv_norm_factor
-
-        self.query_key_value = Linear(
-            self.hidden_size,
-            (config.n_head_kv * 2 + config.n_head) * self.head_dim,
-            bias=config.bias,
-        )
-        self.dense = Linear(self.hidden_size, self.hidden_size, bias=config.bias)
+        if config.new_decoder_architecture:
+            qkv_out_dim = (config.num_kv_heads * 2 + config.num_attention_heads) * self.head_dim
+        elif config.multi_query:
+            qkv_out_dim = self.hidden_size + 2 * self.head_dim
+        else:
+            qkv_out_dim = 3 * self.hidden_size
+        self.query_key_value = FalconLinear(self.hidden_size, qkv_out_dim, bias=config.bias)
+        self.new_decoder_architecture = config.new_decoder_architecture
+        self.multi_query = config.multi_query
+        self.dense = FalconLinear(self.hidden_size, self.hidden_size, bias=config.bias)
         self.attention_dropout = nn.Dropout(config.attention_dropout)
-        self.num_kv = config.n_head_kv
+        self.num_kv_heads = config.num_kv_heads if (self.new_decoder_architecture or not self.multi_query) else 1
 
     def _split_heads(self, fused_qkv: torch.Tensor) -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor]:
         """
-        Split the last dimension into (num_heads, head_dim), results share same memory
-        storage as `fused_qkv`
+        Split the last dimension into (num_heads, head_dim), results share same memory storage as `fused_qkv`
 
         Args:
             fused_qkv (`torch.tensor`, *required*): [batch_size, seq_length, num_heads * 3 * head_dim]
 
         Returns:
-            query: [batch_size, seq_length, num_heads, head_dim]
-            key: [batch_size, seq_length, num_heads, head_dim]
+            query: [batch_size, seq_length, num_heads, head_dim] key: [batch_size, seq_length, num_heads, head_dim]
             value: [batch_size, seq_length, num_heads, head_dim]
         """
-        batch, seq_len, _ = fused_qkv.shape
-        qkv = fused_qkv.view(batch, seq_len, -1, self.num_heads // self.num_kv + 2, 64)
-        q = qkv[:, :, :, :-2]
-        k = qkv[:, :, :, [-2]]
-        v = qkv[:, :, :, [-1]]
-        k = torch.broadcast_to(k, q.shape)
-        v = torch.broadcast_to(v, q.shape)
-
-        q, k, v = [
-            rearrange(
-                x,
-                "batch seq_len group num_heads head_dim ->\
-                batch seq_len (group num_heads) head_dim",
-                head_dim=self.head_dim,
-            )
-            for x in [q, k, v]
-        ]
-        return q, k, v
+        if self.new_decoder_architecture:
+            batch, seq_len, _ = fused_qkv.shape
+            qkv = fused_qkv.view(batch, seq_len, -1, self.num_heads // self.num_kv_heads + 2, self.head_dim)
+            query = qkv[:, :, :, :-2]
+            key = qkv[:, :, :, [-2]]
+            value = qkv[:, :, :, [-1]]
+            key = torch.broadcast_to(key, query.shape)
+            value = torch.broadcast_to(value, query.shape)
+
+            query, key, value = [x.flatten(2, 3) for x in (query, key, value)]
+            return query, key, value
+        elif not self.multi_query:
+            batch_size, seq_length, three_times_hidden_size = fused_qkv.shape
+            fused_qkv = fused_qkv.view(batch_size, seq_length, self.num_heads, 3, self.head_dim)
+            return fused_qkv[..., 0, :], fused_qkv[..., 1, :], fused_qkv[..., 2, :]
+        else:
+            batch_size, seq_length, three_times_hidden_size = fused_qkv.shape
+            fused_qkv = fused_qkv.view(batch_size, seq_length, self.num_heads + 2, self.head_dim)
+            return fused_qkv[..., :-2, :], fused_qkv[..., [-2], :], fused_qkv[..., [-1], :]
 
+    # Copied from transformers.models.bloom.modeling_bloom.BloomAttention._merge_heads
     def _merge_heads(self, x: torch.Tensor) -> torch.Tensor:
         """
         Merge heads together over the last dimenstion
 
         Args:
-            x: (`torch.tensor`, *required*): [batch_size * num_heads, seq_length, head_dim]
+            x (`torch.tensor`, *required*): [batch_size * num_heads, seq_length, head_dim]
 
         Returns:
             torch.tensor: [batch_size, seq_length, num_heads * head_dim]
@@ -242,7 +281,7 @@ class Attention(nn.Module):
     def forward(
         self,
         hidden_states: torch.Tensor,
-        alibi: torch.Tensor,
+        alibi: Optional[torch.Tensor],
         attention_mask: torch.Tensor,
         layer_past: Optional[Tuple[torch.Tensor, torch.Tensor]] = None,
         head_mask: Optional[torch.Tensor] = None,
@@ -250,106 +289,120 @@ class Attention(nn.Module):
         output_attentions: bool = False,
     ):
         fused_qkv = self.query_key_value(hidden_states)  # [batch_size, seq_length, 3 x hidden_size]
-
+        num_kv_heads = self.num_heads if self.new_decoder_architecture else self.num_kv_heads
         # 3 x [batch_size, seq_length, num_heads, head_dim]
         (query_layer, key_layer, value_layer) = self._split_heads(fused_qkv)
 
-        batch_size, q_length, _, _ = query_layer.shape
+        batch_size, query_length, _, _ = query_layer.shape
 
-        query_layer = query_layer.transpose(1, 2).reshape(batch_size * self.num_heads, q_length, self.head_dim)
+        query_layer = query_layer.transpose(1, 2).reshape(batch_size * self.num_heads, query_length, self.head_dim)
         key_layer = key_layer.transpose(1, 2).reshape(
-            batch_size * self.num_heads,
-            q_length,
+            batch_size * num_kv_heads,
+            query_length,
             self.head_dim,
         )
-        value_layer = value_layer.transpose(1, 2).reshape(batch_size * self.num_heads, q_length, self.head_dim)
+        value_layer = value_layer.transpose(1, 2).reshape(batch_size * num_kv_heads, query_length, self.head_dim)
 
-        query_layer, key_layer = self.maybe_rotary(query_layer, key_layer)
+        past_kv_length = 0 if layer_past is None else layer_past[0].shape[1]
+        query_layer, key_layer = self.maybe_rotary(query_layer, key_layer, past_kv_length)
 
         if layer_past is not None:
             past_key, past_value = layer_past
             # concatenate along seq_length dimension:
-            #  - key: [batch_size * self.num_heads, head_dim, kv_length]
+            #  - key: [batch_size * self.num_heads, kv_length, head_dim]
             #  - value: [batch_size * self.num_heads, kv_length, head_dim]
             key_layer = torch.cat((past_key, key_layer), dim=1)
             value_layer = torch.cat((past_value, value_layer), dim=1)
 
         _, kv_length, _ = key_layer.shape
-
-        if use_cache is True:
+        if use_cache:
             present = (key_layer, value_layer)
         else:
             present = None
 
+        attention_mask_float = (attention_mask * 1.0).masked_fill(attention_mask, float("-1e9")).to(query_layer.dtype)
+
+        query_layer_ = query_layer.reshape(batch_size, self.num_heads, -1, self.head_dim)
+        key_layer_ = key_layer.reshape(batch_size, num_kv_heads, -1, self.head_dim)
+        value_layer_ = value_layer.reshape(batch_size, num_kv_heads, -1, self.head_dim)
+
         if alibi is None:
-            query_layer_ = query_layer.reshape(batch_size, self.num_heads, -1, self.head_dim)
-            key_layer_ = key_layer.reshape(batch_size, self.num_heads, -1, self.head_dim)
-            value_layer_ = value_layer.reshape(batch_size, self.num_heads, -1, self.head_dim)
+            if output_attentions:
+                # F.scaled_dot_product_attention doesn't return the attention weights, so we have
+                # to do it by hand if we want them
+                attention_scores = query_layer_ @ key_layer_.transpose(-1, -2)
+                attention_scores /= math.sqrt(self.head_dim)
 
-            attn_output = F.scaled_dot_product_attention(
-                query_layer_, key_layer_, value_layer_, None, 0.0, is_causal=True
-            )
+                attention_scores = F.softmax(
+                    attention_scores + attention_mask_float, dim=-1, dtype=hidden_states.dtype
+                )
+                attn_output = attention_scores @ value_layer_
+            else:
+                attn_output = F.scaled_dot_product_attention(
+                    query_layer_, key_layer_, value_layer_, attention_mask_float, 0.0, is_causal=False
+                )
+                attention_scores = None
 
-            x = attn_output.view(batch_size, self.num_heads, q_length, self.head_dim)
-            x = x.permute(0, 2, 1, 3)
-            attn_output = x.reshape(batch_size, q_length, self.num_heads * self.head_dim)
+            attn_output = attn_output.view(batch_size, self.num_heads, query_length, self.head_dim)
+            attn_output = attn_output.permute(0, 2, 1, 3)
+            attn_output = attn_output.reshape(batch_size, query_length, self.num_heads * self.head_dim)
 
             output_tensor = self.dense(attn_output)
 
-            outputs = (output_tensor, present)
-            assert not output_attentions  # not supported.
-            return outputs
+            if output_attentions:
+                return output_tensor, present, attention_scores
+            else:
+                return output_tensor, present
+
         else:
-            attention_mask_float = (attention_mask * 1.0).masked_fill(attention_mask, -1e9).to(torch.bfloat16)
-            matmul_result = query_layer @ key_layer.transpose(-1, -2)
+            matmul_result = query_layer_ @ key_layer_.transpose(-1, -2)
 
             # change view to [batch_size, num_heads, q_length, kv_length]
-            attention_scores = matmul_result.view(batch_size, self.num_heads, q_length, kv_length)
+            attention_scores = matmul_result.view(batch_size, self.num_heads, query_length, kv_length)
 
             # cast attention scores to fp32, compute scaled softmax and cast back to initial dtype - [batch_size, num_heads, q_length, kv_length]
             input_dtype = attention_scores.dtype
             # `float16` has a minimum value of -65504.0, whereas `bfloat16` and `float32` have a minimum value of `-3.4e+38`
             if input_dtype == torch.float16 or input_dtype == torch.bfloat16:
                 attention_scores = attention_scores.to(torch.float32)
-            # attn_weights = torch.masked_fill(attention_scores, attention_mask, torch.finfo(attention_scores.dtype).min)
-            attention_probs = F.softmax(
-                (attention_scores + alibi.view(batch_size, self.num_heads, 1, -1)) * self.inv_norm_factor
-                + attention_mask_float,
-                dim=-1,
-                dtype=hidden_states.dtype,
-            )
+            # Matt (HF) note: We could possibly use F.scaled_dot_product_attention here too, by
+            # adding (alibi * self.inv_norm_factor) to attention_mask_float. I think this would be mathematically
+            # equivalent and more performant, but there might be a numerical difference. If you're reading this
+            # and you'd like to experiment and maybe file a PR, feel free!
+            attention_logits = attention_scores + alibi.view(batch_size, self.num_heads, 1, -1)
+            attention_logits *= self.inv_norm_factor
+            attention_probs = F.softmax(attention_logits + attention_mask_float, dim=-1, dtype=hidden_states.dtype)
             # [batch_size, num_heads, q_length, kv_length]
             attention_probs = self.attention_dropout(attention_probs)
 
             if head_mask is not None:
                 attention_probs = attention_probs * head_mask
 
-            # change view [batch_size x num_heads, q_length, kv_length]
-            attention_probs_reshaped = attention_probs.view(batch_size * self.num_heads, q_length, kv_length)
+            # change view [batch_size, num_heads, q_length, kv_length]
+            attention_probs_reshaped = attention_probs.view(batch_size, self.num_heads, query_length, kv_length)
 
             # matmul: [batch_size * num_heads, q_length, head_dim]
-            context_layer = attention_probs_reshaped @ value_layer
+            context_layer = (attention_probs_reshaped @ value_layer_).flatten(0, 1)
 
             # change view [batch_size, num_heads, q_length, head_dim]
             context_layer = self._merge_heads(context_layer)
 
             output_tensor = self.dense(context_layer)
 
-            outputs = (output_tensor, present)
             if output_attentions:
-                outputs += (attention_probs,)
-
-            return outputs
+                return output_tensor, present, attention_probs
+            else:
+                return output_tensor, present
 
 
-class MLP(nn.Module):
-    def __init__(self, config: RWConfig):
+class FalconMLP(nn.Module):
+    def __init__(self, config: FalconConfig):
         super().__init__()
         hidden_size = config.hidden_size
 
-        self.dense_h_to_4h = Linear(hidden_size, 4 * hidden_size, bias=config.bias)
+        self.dense_h_to_4h = FalconLinear(hidden_size, 4 * hidden_size, bias=config.bias)
         self.act = nn.GELU()
-        self.dense_4h_to_h = Linear(4 * hidden_size, hidden_size, bias=config.bias)
+        self.dense_4h_to_h = FalconLinear(4 * hidden_size, hidden_size, bias=config.bias)
         self.hidden_dropout = config.hidden_dropout
 
     def forward(self, x: torch.Tensor) -> torch.Tensor:
@@ -358,43 +411,47 @@ class MLP(nn.Module):
         return x
 
 
-class DecoderLayer(nn.Module):
-    def __init__(self, config: RWConfig):
+class FalconDecoderLayer(nn.Module):
+    def __init__(self, config: FalconConfig):
         super().__init__()
         hidden_size = config.hidden_size
-
-        self.ln_attn = LayerNorm(hidden_size, eps=config.layer_norm_epsilon)
-        self.ln_mlp = LayerNorm(hidden_size, eps=config.layer_norm_epsilon)
-
-        self.num_heads = config.n_head
-        self.self_attention = Attention(config)
-
-        self.mlp = MLP(config)
-
-        self.apply_residual_connection_post_layernorm = config.apply_residual_connection_post_layernorm
+        self.num_heads = config.num_attention_heads
+        self.self_attention = FalconAttention(config)
+        self.mlp = FalconMLP(config)
         self.hidden_dropout = config.hidden_dropout
-
         self.config = config
 
+        if config.new_decoder_architecture:
+            # The layer norm before self-attention
+            self.ln_attn = LayerNorm(hidden_size, eps=config.layer_norm_epsilon)
+            # The layer norm before the MLP
+            self.ln_mlp = LayerNorm(hidden_size, eps=config.layer_norm_epsilon)
+        else:
+            self.input_layernorm = LayerNorm(hidden_size, eps=config.layer_norm_epsilon)
+            if not config.parallel_attn:
+                self.post_attention_layernorm = LayerNorm(hidden_size, eps=config.layer_norm_epsilon)
+
     def forward(
         self,
         hidden_states: torch.Tensor,
-        alibi: torch.Tensor,
+        alibi: Optional[torch.Tensor],
         attention_mask: torch.Tensor,
         layer_past: Optional[Tuple[torch.Tensor, torch.Tensor]] = None,
         head_mask: Optional[torch.Tensor] = None,
         use_cache: bool = False,
         output_attentions: bool = False,
     ):
-
-        ln_attn = self.ln_attn(hidden_states)
-        ln_mlp = self.ln_mlp(hidden_states)
-
         residual = hidden_states
 
+        if self.config.new_decoder_architecture:
+            attention_layernorm_out = self.ln_attn(hidden_states)
+            mlp_layernorm_out = self.ln_mlp(hidden_states)
+        else:
+            attention_layernorm_out = self.input_layernorm(hidden_states)
+
         # Self attention.
         attn_outputs = self.self_attention(
-            ln_attn,
+            attention_layernorm_out,
             layer_past=layer_past,
             attention_mask=attention_mask,
             alibi=alibi,
@@ -405,14 +462,24 @@ class DecoderLayer(nn.Module):
 
         attention_output = attn_outputs[0]
 
+        if not self.config.new_decoder_architecture:
+            if self.config.parallel_attn:
+                mlp_layernorm_out = attention_layernorm_out
+            else:
+                residual = dropout_add(
+                    attention_output, residual, self.config.attention_dropout, training=self.training
+                )
+                mlp_layernorm_out = self.post_attention_layernorm(residual)
+
         outputs = attn_outputs[1:]
 
         # MLP.
-        mlp_output = self.mlp(ln_mlp)
+        mlp_output = self.mlp(mlp_layernorm_out)
 
-        output = dropout_add(
-            mlp_output + attention_output, residual, self.config.hidden_dropout, training=self.training
-        )
+        if self.config.new_decoder_architecture or self.config.parallel_attn:
+            mlp_output += attention_output
+
+        output = dropout_add(mlp_output, residual, self.config.hidden_dropout, training=self.training)
 
         if use_cache:
             outputs = (output,) + outputs
@@ -422,24 +489,93 @@ class DecoderLayer(nn.Module):
         return outputs  # hidden_states, present, attentions
 
 
-class RWPreTrainedModel(PreTrainedModel):
-    _keys_to_ignore_on_load_missing = [r"h.*.self_attention.scale_mask_softmax.causal_mask", r"lm_head.weight"]
+FALCON_START_DOCSTRING = r"""
+
+    This model inherits from [`PreTrainedModel`]. Check the superclass documentation for the generic methods the
+    library implements for all its model (such as downloading or saving, resizing the input embeddings etc.)
+
+    This model is also a PyTorch [torch.nn.Module](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) subclass.
+    Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage
+    and behavior.
+
+    Parameters:
+        config ([`FalconConfig`]): Model configuration class with all the parameters of the model.
+            Initializing with a config file does not load the weights associated with the model, only the
+            configuration. Check out the [`~PreTrainedModel.from_pretrained`] method to load the model weights.
+"""
+
+FALCON_INPUTS_DOCSTRING = r"""
+    Args:
+        input_ids (`torch.LongTensor` of shape `(batch_size, input_ids_length)`):
+            `input_ids_length` = `sequence_length` if `past_key_values` is `None` else `past_key_values[0][0].shape[2]`
+            (`sequence_length` of input past key value states). Indices of input sequence tokens in the vocabulary.
+
+            If `past_key_values` is used, only `input_ids` that do not have their past calculated should be passed as
+            `input_ids`.
+
+            Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and
+            [`PreTrainedTokenizer.__call__`] for details.
+
+            [What are input IDs?](../glossary#input-ids)
+        past_key_values (`Tuple[Tuple[torch.Tensor]]` of length `config.num_hidden_layers`):
+            Contains precomputed hidden-states (key and values in the attention blocks) as computed by the model (see
+            `past_key_values` output below). Can be used to speed up sequential decoding. The `input_ids` which have
+            their past given to this model should not be passed as `input_ids` as they have already been computed.
+
+            Each element of `past_key_values` is a tuple (past_key, past_value):
+            - past_key: [batch_size * num_heads, head_dim, kv_length]
+            - past_value: [batch_size * num_heads, kv_length, head_dim]
+        attention_mask (`torch.FloatTensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Mask to avoid performing attention on padding token indices. Mask values selected in `[0, 1]`:
+
+            - 1 for tokens that are **not masked**,
+            - 0 for tokens that are **masked**.
+
+            [What are attention masks?](../glossary#attention-mask)
+        head_mask (`torch.FloatTensor` of shape `(num_heads,)` or `(num_layers, num_heads)`, *optional*):
+            Mask to nullify selected heads of the self-attention modules. Mask values selected in `[0, 1]`:
+
+            - 1 indicates the head is **not masked**,
+            - 0 indicates the head is **masked**.
+
+        inputs_embeds (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`, *optional*):
+            Optionally, instead of passing `input_ids` you can choose to directly pass an embedded representation. This
+            is useful if you want more control over how to convert `input_ids` indices into associated vectors than the
+            model's internal embedding lookup matrix.
+
+            If `past_key_values` is used, optionally only the last `inputs_embeds` have to be input (see
+            `past_key_values`).
+        use_cache (`bool`, *optional*):
+            If set to `True`, `past_key_values` key value states are returned and can be used to speed up decoding (see
+            `past_key_values`).
+        output_attentions (`bool`, *optional*):
+            Whether or not to return the attentions tensors of all attention layers. See `attentions` under returned
+            tensors for more detail.
+        output_hidden_states (`bool`, *optional*):
+            Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors for
+            more detail.
+        return_dict (`bool`, *optional*):
+            Whether or not to return a [`~file_utils.ModelOutput`] instead of a plain tuple.
+"""
+
+
+class FalconPreTrainedModel(PreTrainedModel):
     """
     An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained
     models.
     """
 
-    config_class = RWConfig
+    config_class = FalconConfig
     base_model_prefix = "transformer"
     supports_gradient_checkpointing = True
-    _no_split_modules = ["DecoderLayer"]
+    _no_split_modules = ["FalconDecoderLayer"]
 
     def __init__(self, *inputs, **kwargs):
         super().__init__(*inputs, **kwargs)
 
     def _init_weights(self, module: nn.Module):
         """Initialize the weights."""
-        if isinstance(module, nn.Linear) or isinstance(module, Linear):
+        if isinstance(module, nn.Linear) or isinstance(module, FalconLinear):
             # 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)
@@ -453,26 +589,28 @@ class RWPreTrainedModel(PreTrainedModel):
             module.bias.data.zero_()
             module.weight.data.fill_(1.0)
 
+    # Copied from transformers.models.bloom.modeling_bloom.BloomPreTrainedModel._set_gradient_checkpointing with BloomModel->FalconModel
     def _set_gradient_checkpointing(self, module: nn.Module, value: bool = False):
-        if isinstance(module, RWModel):
+        if isinstance(module, FalconModel):
             module.gradient_checkpointing = value
 
     @staticmethod
-    def _convert_to_standard_cache(
+    def _convert_cache_to_standard_format(
         past_key_value: Tuple[Tuple[torch.Tensor, torch.Tensor]], batch_size: int
     ) -> Tuple[Tuple[torch.Tensor, torch.Tensor]]:
         """
         Standardizes the format of the cache so as to match most implementations, i.e. to tuple(tuple([batch_size,
         num_heads, ...]))
         """
-        batch_size_times_num_heads, head_dim, seq_length = past_key_value[0][0].shape
+        batch_size_times_num_heads, kv_length, head_dim = past_key_value[0][0].shape
+        # [batch_size * self.num_heads, kv_length, head_dim] -> [batch_size, num_heads, kv_length, head_dim]
+        # Note that don't want to use self.num_attention_heads because the number of heads may vary depending
+        # on whether we use multi_query attention.
         num_heads = batch_size_times_num_heads // batch_size
-        # key: [batch_size * num_heads, head_dim, seq_length] -> [batch_size, num_heads, head_dim, seq_length]
-        # value: [batch_size * num_heads, seq_length, head_dim] -> [batch_size, num_heads, seq_length, head_dim]
         return tuple(
             (
-                layer_past[0].view(batch_size, num_heads, head_dim, seq_length),
-                layer_past[1].view(batch_size, num_heads, seq_length, head_dim),
+                layer_past[0].view(batch_size, num_heads, kv_length, head_dim),
+                layer_past[1].view(batch_size, num_heads, kv_length, head_dim),
             )
             for layer_past in past_key_value
         )
@@ -481,32 +619,35 @@ class RWPreTrainedModel(PreTrainedModel):
     def _convert_to_rw_cache(
         past_key_value: Tuple[Tuple[torch.Tensor, torch.Tensor]]
     ) -> Tuple[Tuple[torch.Tensor, torch.Tensor]]:
-        batch_size, num_heads, head_dim, seq_length = past_key_value[0][0].shape
+        batch_size, num_heads, kv_length, head_dim = past_key_value[0][0].shape
         batch_size_times_num_heads = batch_size * num_heads
-        # key:  [batch_size, num_heads, head_dim, seq_length] -> [batch_size * num_heads, head_dim, seq_length]
-        # value: [batch_size, num_heads, seq_length, head_dim] -> [batch_size * num_heads, seq_length, head_dim]
+        # [batch_size, num_heads, kv_length, head_dim] -> [batch_size * num_heads, kv_length, head_dim]
         return tuple(
             (
-                layer_past[0].view(batch_size_times_num_heads, head_dim, seq_length),
-                layer_past[1].view(batch_size_times_num_heads, seq_length, head_dim),
+                layer_past[0].view(batch_size_times_num_heads, kv_length, head_dim),
+                layer_past[1].view(batch_size_times_num_heads, kv_length, head_dim),
             )
             for layer_past in past_key_value
         )
 
 
-class RWModel(RWPreTrainedModel):
-    def __init__(self, config: RWConfig):
+@add_start_docstrings(
+    "The bare Falcon Model transformer outputting raw hidden-states without any specific head on top.",
+    FALCON_START_DOCSTRING,
+)
+class FalconModel(FalconPreTrainedModel):
+    def __init__(self, config: FalconConfig):
         super().__init__(config)
 
         self.embed_dim = config.hidden_size
-        self.num_heads = config.n_head
-        self.alibi = config.alibi
+        self.num_heads = config.num_attention_heads
+        self.use_alibi = config.alibi
 
         # Embedding + LN Embedding
         self.word_embeddings = nn.Embedding(config.vocab_size, self.embed_dim)
 
         # Transformer blocks
-        self.h = nn.ModuleList([DecoderLayer(config) for _ in range(config.num_hidden_layers)])
+        self.h = nn.ModuleList([FalconDecoderLayer(config) for _ in range(config.num_hidden_layers)])
 
         # Final Layer Norm
         self.ln_f = LayerNorm(self.embed_dim, eps=config.layer_norm_epsilon)
@@ -519,22 +660,31 @@ class RWModel(RWPreTrainedModel):
     def get_input_embeddings(self):
         return self.word_embeddings
 
+    @staticmethod
     def _prepare_attn_mask(
-        self, attention_mask: torch.Tensor, input_shape: Tuple[int, int], past_key_values_length: int
+        attention_mask: torch.Tensor, input_shape: Tuple[int, int], past_key_values_length: int
     ) -> torch.BoolTensor:
-        # create causal mask
-        # [batch_size, seq_length] -> [batch_size, 1, tgt_length, src_length]
+        # Create a causal mask
+        # The attention mask we receive as input should cover the whole extended sequence, including any past
+        # cache, so its shape should be [batch_size, seq_length + past_key_values_length]
+        # The output shape will be [batch_size, 1, seq_length, seq_length + past_key_values_length]
+        if input_shape[1] + past_key_values_length != attention_mask.shape[1]:
+            raise ValueError(
+                "Attention mask shape should be (batch_size, seq_length + past_key_values_length)"
+                f" but is {attention_mask.shape} with input_ids shape {input_shape} and past length"
+                f" {past_key_values_length}."
+            )
         combined_attention_mask = None
         device = attention_mask.device
-        _, src_length = input_shape
+        _, seq_length = input_shape
 
-        if src_length > 1:
+        if seq_length > 1:
             combined_attention_mask = _make_causal_mask(
                 input_shape, device=device, past_key_values_length=past_key_values_length
             )
 
-        # [batch_size, seq_length] -> [batch_size, 1, tgt_length, src_length]
-        expanded_attn_mask = _expand_mask(attention_mask, tgt_length=src_length)
+        # [batch_size, seq_length + past_key_values_length] -> [batch_size, 1, seq_length, seq_length + past_key_values_length]
+        expanded_attn_mask = _expand_mask(attention_mask, past_key_values_length=past_key_values_length)
         combined_attention_mask = (
             expanded_attn_mask if combined_attention_mask is None else expanded_attn_mask | combined_attention_mask
         )
@@ -544,6 +694,12 @@ class RWModel(RWPreTrainedModel):
     def set_input_embeddings(self, new_embeddings: torch.Tensor):
         self.word_embeddings = new_embeddings
 
+    @add_start_docstrings_to_model_forward(FALCON_INPUTS_DOCSTRING)
+    @add_code_sample_docstrings(
+        checkpoint=_CHECKPOINT_FOR_DOC,
+        output_type=BaseModelOutputWithPastAndCrossAttentions,
+        config_class=_CONFIG_FOR_DOC,
+    )
     def forward(
         self,
         input_ids: Optional[torch.LongTensor] = None,
@@ -555,18 +711,7 @@ class RWModel(RWPreTrainedModel):
         output_attentions: Optional[bool] = None,
         output_hidden_states: Optional[bool] = None,
         return_dict: Optional[bool] = None,
-        **deprecated_arguments,
     ) -> Union[Tuple[torch.Tensor, ...], BaseModelOutputWithPastAndCrossAttentions]:
-        if deprecated_arguments.pop("position_ids", False) is not False:
-            # `position_ids` could have been `torch.Tensor` or `None` so defaulting pop to `False` allows to detect if users were passing explicitly `None`
-            warnings.warn(
-                "`position_ids` have no functionality in BLOOM and will be removed in v5.0.0. You can safely ignore"
-                " passing `position_ids`.",
-                FutureWarning,
-            )
-        if len(deprecated_arguments) > 0:
-            raise ValueError(f"Got unexpected arguments: {deprecated_arguments}")
-
         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
@@ -585,12 +730,14 @@ class RWModel(RWPreTrainedModel):
 
         if past_key_values is None:
             past_key_values = tuple([None] * len(self.h))
+        else:
+            past_key_values = self._convert_to_rw_cache(past_key_values)
 
         # Prepare head mask if needed
         # 1.0 in head_mask indicate we keep the head
         # attention_probs has shape batch_size x num_heads x N x N
         # head_mask has shape n_layer x batch x num_heads x N x N
-        head_mask = self.get_head_mask(head_mask, self.config.n_layer)
+        head_mask = self.get_head_mask(head_mask, self.config.num_hidden_layers)
 
         if inputs_embeds is None:
             inputs_embeds = self.word_embeddings(input_ids)
@@ -602,17 +749,15 @@ class RWModel(RWPreTrainedModel):
         all_hidden_states = () if output_hidden_states else None
 
         # Compute alibi tensor: check build_alibi_tensor documentation
-        seq_length_with_past = seq_length
         past_key_values_length = 0
         if past_key_values[0] is not None:
-            past_key_values_length = past_key_values[0][0].shape[2]
-            seq_length_with_past = seq_length_with_past + past_key_values_length
+            past_key_values_length = past_key_values[0][0].shape[1]  # 1 because RW-cache, not standard format
         if attention_mask is None:
-            attention_mask = torch.ones((batch_size, seq_length_with_past), device=hidden_states.device)
+            attention_mask = torch.ones((batch_size, seq_length + past_key_values_length), device=hidden_states.device)
         else:
             attention_mask = attention_mask.to(hidden_states.device)
 
-        if self.alibi:
+        if self.use_alibi:
             alibi = build_alibi_tensor(attention_mask, self.num_heads, dtype=hidden_states.dtype)
         else:
             alibi = None
@@ -624,12 +769,10 @@ class RWModel(RWPreTrainedModel):
         )
 
         for i, (block, layer_past) in enumerate(zip(self.h, past_key_values)):
-
             if output_hidden_states:
                 all_hidden_states = all_hidden_states + (hidden_states,)
 
             if self.gradient_checkpointing and self.training:
-
                 if use_cache:
                     logger.warning(
                         "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
@@ -674,6 +817,9 @@ class RWModel(RWPreTrainedModel):
         if output_hidden_states:
             all_hidden_states = all_hidden_states + (hidden_states,)
 
+        if presents is not None:
+            presents = self._convert_cache_to_standard_format(presents, batch_size)
+
         if not return_dict:
             return tuple(v for v in [hidden_states, presents, all_hidden_states, all_self_attentions] if v is not None)
 
@@ -685,12 +831,16 @@ class RWModel(RWPreTrainedModel):
         )
 
 
-class RWForCausalLM(RWPreTrainedModel):
-    _keys_to_ignore_on_load_missing = [r"h.*.self_attention.scale_mask_softmax.causal_mask", r"lm_head.weight"]
+@add_start_docstrings(
+    "The Falcon Model transformer with a language modeling head on top (linear layer with weights tied to the input embeddings).",
+    FALCON_START_DOCSTRING,
+)
+class FalconForCausalLM(FalconPreTrainedModel):
+    _tied_weights_keys = ["lm_head.weight"]
 
-    def __init__(self, config: RWConfig):
+    def __init__(self, config: FalconConfig):
         super().__init__(config)
-        self.transformer = RWModel(config)
+        self.transformer = FalconModel(config)
         self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
 
         # Initialize weights and apply final processing
@@ -705,25 +855,26 @@ class RWForCausalLM(RWPreTrainedModel):
     def prepare_inputs_for_generation(
         self,
         input_ids: torch.LongTensor,
-        past: Optional[torch.Tensor] = None,
+        past_key_values: Optional[torch.Tensor] = None,
         attention_mask: Optional[torch.Tensor] = None,
         **kwargs,
     ) -> dict:
-        # only last token for input_ids if past is not None
-        if past:
-            input_ids = input_ids[:, -1].unsqueeze(-1)
-
-            # the cache may be in the stardard format (e.g. in contrastive search), convert to our's format if needed
-            if past[0][0].shape[0] == input_ids.shape[0]:
-                past = self._convert_to_rw_cache(past)
+        if past_key_values is not None:
+            input_ids = input_ids[:, -1:]
 
         return {
             "input_ids": input_ids,
-            "past_key_values": past,
+            "past_key_values": past_key_values,
             "use_cache": kwargs.get("use_cache"),
             "attention_mask": attention_mask,
         }
 
+    @add_start_docstrings_to_model_forward(FALCON_INPUTS_DOCSTRING)
+    @add_code_sample_docstrings(
+        checkpoint=_CHECKPOINT_FOR_DOC,
+        output_type=CausalLMOutputWithCrossAttentions,
+        config_class=_CONFIG_FOR_DOC,
+    )
     def forward(
         self,
         input_ids: Optional[torch.LongTensor] = None,
@@ -736,7 +887,6 @@ class RWForCausalLM(RWPreTrainedModel):
         output_attentions: Optional[bool] = None,
         output_hidden_states: Optional[bool] = None,
         return_dict: Optional[bool] = None,
-        **deprecated_arguments,
     ) -> Union[Tuple[torch.Tensor], CausalLMOutputWithCrossAttentions]:
         r"""
         labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
@@ -744,15 +894,6 @@ class RWForCausalLM(RWPreTrainedModel):
             `labels = input_ids` Indices are selected in `[-100, 0, ..., config.vocab_size]` All labels set to `-100`
             are ignored (masked), the loss is only computed for labels in `[0, ..., config.vocab_size]`
         """
-        if deprecated_arguments.pop("position_ids", False) is not False:
-            # `position_ids` could have been `torch.Tensor` or `None` so defaulting pop to `False` allows to detect if users were passing explicitly `None`
-            warnings.warn(
-                "`position_ids` have no functionality in BLOOM and will be removed in v5.0.0. You can safely ignore"
-                " passing `position_ids`.",
-                FutureWarning,
-            )
-        if len(deprecated_arguments) > 0:
-            raise ValueError(f"Got unexpected arguments: {deprecated_arguments}")
 
         return_dict = return_dict if return_dict is not None else self.config.use_return_dict
 
@@ -805,7 +946,6 @@ class RWForCausalLM(RWPreTrainedModel):
 
         Output shares the same memory storage as `past`.
         """
-        standardized_past = self._convert_to_standard_cache(past, batch_size=len(beam_idx))
 
         # Get a copy of `beam_idx` on all the devices where we need those indices.
         device_to_beam_idx = {
@@ -816,23 +956,42 @@ class RWForCausalLM(RWPreTrainedModel):
                 layer_past[0].index_select(0, device_to_beam_idx[layer_past[0].device]),
                 layer_past[1].index_select(0, device_to_beam_idx[layer_past[0].device]),
             )
-            for layer_past in standardized_past
+            for layer_past in past
         )
-        return self._convert_to_rw_cache(reordered_past)
+        return reordered_past
 
 
-class RWForSequenceClassification(RWPreTrainedModel):
-    _keys_to_ignore_on_load_missing = [r"h.*.self_attention.scale_mask_softmax.causal_mask", r"lm_head.weight"]
-
-    def __init__(self, config: RWConfig):
+@add_start_docstrings(
+    """
+    The Falcon Model transformer with a sequence classification head on top (linear layer).
+
+    [`FalconForSequenceClassification`] uses the last token in order to do the classification, as other causal models
+    (e.g. GPT-1) do.
+
+    Since it does classification on the last token, it requires to know the position of the last token. If a
+    `pad_token_id` is defined in the configuration, it finds the last token that is not a padding token in each row. If
+    no `pad_token_id` is defined, it simply takes the last value in each row of the batch. Since it cannot guess the
+    padding tokens when `inputs_embeds` are passed instead of `input_ids`, it does the same (take the last value in
+    each row of the batch).
+    """,
+    FALCON_START_DOCSTRING,
+)
+class FalconForSequenceClassification(FalconPreTrainedModel):
+    def __init__(self, config: FalconConfig):
         super().__init__(config)
         self.num_labels = config.num_labels
-        self.transformer = RWModel(config)
+        self.transformer = FalconModel(config)
         self.score = nn.Linear(config.hidden_size, config.num_labels, bias=False)
 
         # Initialize weights and apply final processing
         self.post_init()
 
+    @add_start_docstrings_to_model_forward(FALCON_INPUTS_DOCSTRING)
+    @add_code_sample_docstrings(
+        checkpoint=_CHECKPOINT_FOR_DOC,
+        output_type=SequenceClassifierOutputWithPast,
+        config_class=_CONFIG_FOR_DOC,
+    )
     def forward(
         self,
         input_ids: Optional[torch.LongTensor] = None,
@@ -845,7 +1004,6 @@ class RWForSequenceClassification(RWPreTrainedModel):
         output_attentions: Optional[bool] = None,
         output_hidden_states: Optional[bool] = None,
         return_dict: Optional[bool] = None,
-        **deprecated_arguments,
     ) -> Union[Tuple[torch.Tensor], SequenceClassifierOutputWithPast]:
         r"""
         labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
@@ -853,15 +1011,6 @@ class RWForSequenceClassification(RWPreTrainedModel):
             config.num_labels - 1]`. If `config.num_labels == 1` a regression loss is computed (Mean-Square loss), If
             `config.num_labels > 1` a classification loss is computed (Cross-Entropy).
         """
-        if deprecated_arguments.pop("position_ids", False) is not False:
-            # `position_ids` could have been `torch.Tensor` or `None` so defaulting pop to `False` allows to detect if users were passing explicitly `None`
-            warnings.warn(
-                "`position_ids` have no functionality in BLOOM and will be removed in v5.0.0. You can safely ignore"
-                " passing `position_ids`.",
-                FutureWarning,
-            )
-        if len(deprecated_arguments) > 0:
-            raise ValueError(f"Got unexpected arguments: {deprecated_arguments}")
 
         return_dict = return_dict if return_dict is not None else self.config.use_return_dict
 
@@ -936,17 +1085,22 @@ class RWForSequenceClassification(RWPreTrainedModel):
         )
 
 
-class RWForTokenClassification(RWPreTrainedModel):
-    _keys_to_ignore_on_load_missing = [r"h.*.self_attention.scale_mask_softmax.causal_mask", r"lm_head.weight"]
-
-    def __init__(self, config: RWConfig):
+@add_start_docstrings(
+    """
+    Falcon Model with a token classification head on top (a linear layer on top of the hidden-states output) e.g. for
+    Named-Entity-Recognition (NER) tasks.
+    """,
+    FALCON_START_DOCSTRING,
+)
+class FalconForTokenClassification(FalconPreTrainedModel):
+    def __init__(self, config: FalconConfig):
         super().__init__(config)
         self.num_labels = config.num_labels
 
-        self.transformer = RWModel(config)
-        if hasattr(config, "classifier_dropout") and config.classifier_dropout is not None:
+        self.transformer = FalconModel(config)
+        if getattr(config, "classifier_dropout", None) is not None:
             classifier_dropout = config.classifier_dropout
-        elif hasattr(config, "hidden_dropout") and config.hidden_dropout is not None:
+        elif getattr(config, "hidden_dropout", None) is not None:
             classifier_dropout = config.hidden_dropout
         else:
             classifier_dropout = 0.1
@@ -956,6 +1110,12 @@ class RWForTokenClassification(RWPreTrainedModel):
         # Initialize weights and apply final processing
         self.post_init()
 
+    @add_start_docstrings_to_model_forward(FALCON_INPUTS_DOCSTRING)
+    @add_code_sample_docstrings(
+        checkpoint=_CHECKPOINT_FOR_DOC,
+        output_type=TokenClassifierOutput,
+        config_class=_CONFIG_FOR_DOC,
+    )
     def forward(
         self,
         input_ids: Optional[torch.LongTensor] = None,
@@ -968,7 +1128,6 @@ class RWForTokenClassification(RWPreTrainedModel):
         output_attentions: Optional[bool] = None,
         output_hidden_states: Optional[bool] = None,
         return_dict: Optional[bool] = None,
-        **deprecated_arguments,
     ) -> Union[Tuple[torch.Tensor], TokenClassifierOutput]:
         r"""
         labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
@@ -976,15 +1135,6 @@ class RWForTokenClassification(RWPreTrainedModel):
             config.num_labels - 1]`. If `config.num_labels == 1` a regression loss is computed (Mean-Square loss), If
             `config.num_labels > 1` a classification loss is computed (Cross-Entropy).
         """
-        if deprecated_arguments.pop("position_ids", False) is not False:
-            # `position_ids` could have been `torch.Tensor` or `None` so defaulting pop to `False` allows to detect if users were passing explicitly `None`
-            warnings.warn(
-                "`position_ids` have no functionality in BLOOM and will be removed in v5.0.0. You can safely ignore"
-                " passing `position_ids`.",
-                FutureWarning,
-            )
-        if len(deprecated_arguments) > 0:
-            raise ValueError(f"Got unexpected arguments: {deprecated_arguments}")
 
         return_dict = return_dict if return_dict is not None else self.config.use_return_dict
 
@@ -1008,7 +1158,9 @@ class RWForTokenClassification(RWPreTrainedModel):
         if labels is not None:
             batch_size, seq_length = labels.shape
             loss_fct = CrossEntropyLoss()
-            loss = loss_fct(logits.view(batch_size * seq_length, self.num_labels), labels.view(batch_size * seq_length))
+            loss = loss_fct(
+                logits.view(batch_size * seq_length, self.num_labels), labels.view(batch_size * seq_length)
+            )
 
         if not return_dict:
             output = (logits,) + transformer_outputs[2:]
@@ -1022,22 +1174,27 @@ class RWForTokenClassification(RWPreTrainedModel):
         )
 
 
-class RWForQuestionAnswering(RWPreTrainedModel):
-    _keys_to_ignore_on_load_missing = [r"h.*.self_attention.scale_mask_softmax.causal_mask", r"lm_head.weight"]
-
+@add_start_docstrings(
+    """
+    The Falcon Model transformer with a span classification head on top for extractive question-answering tasks like
+    SQuAD (a linear layers on top of the hidden-states output to compute `span start logits` and `span end logits`).
+    """,
+    FALCON_START_DOCSTRING,
+)
+class FalconForQuestionAnswering(FalconPreTrainedModel):
     def __init__(self, config):
         super().__init__(config)
-        self.transformer = RWModel(config)
+        self.transformer = FalconModel(config)
         self.qa_outputs = nn.Linear(config.hidden_size, 2)
 
         # Initialize weights and apply final processing
         self.post_init()
 
+    @add_start_docstrings_to_model_forward(FALCON_INPUTS_DOCSTRING)
     def forward(
         self,
         input_ids: Optional[torch.LongTensor] = None,
         attention_mask: Optional[torch.FloatTensor] = None,
-        position_ids: Optional[torch.LongTensor] = None,
         head_mask: Optional[torch.FloatTensor] = None,
         inputs_embeds: Optional[torch.FloatTensor] = None,
         start_positions: Optional[torch.LongTensor] = None,
@@ -1061,7 +1218,6 @@ class RWForQuestionAnswering(RWPreTrainedModel):
         outputs = self.transformer(
             input_ids,
             attention_mask=attention_mask,
-            position_ids=position_ids,
             head_mask=head_mask,
             inputs_embeds=inputs_embeds,
             output_attentions=output_attentions,

tokenizer_config.json

@@ -1,8 +1,12 @@
 {
   "add_prefix_space": false,
   "eos_token": "<|endoftext|>",
+  "model_input_names": [
+    "input_ids",
+    "attention_mask"
+  ],
   "model_max_length": 2048,
   "name_or_path": "tiiuae/falcon_tokenizer",
   "special_tokens_map_file": null,
   "tokenizer_class": "PreTrainedTokenizerFast"
-}
+}