llama

llama/__init__.py

@@ -0,0 +1,57 @@
+# Copyright 2020 The HuggingFace 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.
+from typing import TYPE_CHECKING
+
+from transformers.utils import OptionalDependencyNotAvailable, _LazyModule, is_torch_available
+
+
+_import_structure = {
+    "configuration_llama": ["LLAMA_PRETRAINED_CONFIG_ARCHIVE_MAP", "LLaMAConfig"],
+    "tokenization_llama": ["LLaMATokenizer"],
+}
+
+try:
+    if not is_torch_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
+    _import_structure["modeling_llama"] = [
+        "LLaMAForCausalLM",
+        "LLaMAModel",
+        "LLaMAPreTrainedModel",
+    ]
+
+
+if TYPE_CHECKING:
+    from .configuration_llama import LLAMA_PRETRAINED_CONFIG_ARCHIVE_MAP, LLaMAConfig
+    from .tokenization_llama import LLaMATokenizer
+
+    try:
+        if not is_torch_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
+        from .modeling_llama import (
+            LLaMAForCausalLM,
+            LLaMAModel,
+            LLaMAPreTrainedModel,
+        )
+
+
+else:
+    import sys
+
+    sys.modules[__name__] = _LazyModule(__name__, globals()["__file__"], _import_structure, module_spec=__spec__)

llama/configuration_llama.py

@@ -0,0 +1,111 @@
+# coding=utf-8
+# Copyright 2022 The FAIR team of Meta AI and The 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.
+""" LLaMA model configuration"""
+
+from transformers.configuration_utils import PretrainedConfig
+from transformers.utils import logging
+
+
+logger = logging.get_logger(__name__)
+
+LLAMA_PRETRAINED_CONFIG_ARCHIVE_MAP = {}
+
+
+class LLaMAConfig(PretrainedConfig):
+    r"""
+    This is the configuration class to store the configuration of a [`~LLaMAModel`]. It is used to instantiate an LLaMA
+    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 LLaMA-7B.
+
+    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 32000):
+            Vocabulary size of the LLaMA model. Defines the number of different tokens that can be represented by the
+            `inputs_ids` passed when calling [`~LLaMAModel`] or [`~TFLLaMAModel`].
+        hidden_size (`int`, *optional*, defaults to 4096):
+            Dimension of the hidden representations.
+        intermediate_size (`int`, *optional*, defaults to 11008):
+            Dimension of the MLP representations.
+        num_hidden_layers (`int`, *optional*, defaults to 32):
+            Number of hidden layers in the Transformer encoder.
+        num_attention_heads (`int`, *optional*, defaults to 32):
+            Number of attention heads for each attention layer in the Transformer encoder.
+        hidden_act (`str` or `function`, *optional*, defaults to `"silu"`):
+            The non-linear activation function (function or string) in the decoder.
+        max_sequence_length (`int`, *optional*, defaults to 2048):
+            Max sequence length for model (for RoPE computation)
+        initializer_range (`float`, *optional*, defaults to 0.02):
+            The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
+        rms_norm_eps (`float`, *optional*, defaults to 1e-12):
+            The epsilon used by the rms normalization layers.
+        use_cache (`bool`, *optional*, defaults to `True`):
+            Whether or not the model should return the last key/values attentions (not used by all models). Only
+            relevant if `config.is_decoder=True`.
+        tie_word_embeddings(`bool`, *optional*, defaults to `False`):
+            Whether to tie weight embeddings
+        Example:
+
+    ```python
+    >>> from llama import LLaMAModel, LLaMAConfig
+
+    >>> # Initializing a LLaMA llama-7b style configuration
+    >>> configuration = LLaMAConfig()
+
+    >>> # Initializing a model from the llama-7b style configuration
+    >>> model = LLaMAModel(configuration)
+
+    >>> # Accessing the model configuration
+    >>> configuration = model.config
+    ```"""
+    model_type = "llama"
+
+    def __init__(
+        self,
+        vocab_size=32000,
+        hidden_size=4096,
+        intermediate_size=11008,
+        num_hidden_layers=32,
+        num_attention_heads=32,
+        hidden_act="silu",
+        max_sequence_length=2048,
+        initializer_range=0.02,
+        rms_norm_eps=1e-6,
+        use_cache=True,
+        pad_token_id=-1,
+        bos_token_id=0,
+        eos_token_id=1,
+        tie_word_embeddings=False,
+        **kwargs,
+    ):
+        self.vocab_size = vocab_size
+        self.hidden_size = hidden_size
+        self.intermediate_size = intermediate_size
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+        self.hidden_act = hidden_act
+        self.max_sequence_length = max_sequence_length
+        self.initializer_range = initializer_range
+        self.rms_norm_eps = rms_norm_eps
+        self.use_cache = use_cache
+        super().__init__(
+            pad_token_id=pad_token_id,
+            bos_token_id=bos_token_id,
+            eos_token_id=eos_token_id,
+            tie_word_embeddings=tie_word_embeddings,
+            **kwargs,
+        )

llama/convert_llama_weights_to_hf.py

@@ -0,0 +1,264 @@
+import argparse
+import json
+import os
+import shutil
+
+import torch
+
+
+"""
+Sample usage:
+
+    ```
+    python src/transformers/models/llama/convert_llama_weights_to_hf.py \
+        --input_dir /path/to/downloaded/llama/weights --model_size 7B --output_dir /output/path
+    ```
+
+Thereafter, models can be loaded via:
+
+    ```
+    tokenizer = transformers.LLaMATokenizer.from_pretrained("/output/path/tokenizer/")
+
+    model = transformers.LLaMAForCausalLM.from_pretrained("/output/path/llama-7b/")
+    ```
+"""
+
+INTERMEDIATE_SIZE_MAP = {
+    "7B": 11008,
+    "13B": 13824,
+    "30B": 17920,
+    "65B": 22016,
+}
+NUM_SHARDS = {
+    "7B": 1,
+    "13B": 2,
+    "30B": 4,
+    "65B": 8,
+}
+
+
+def read_json(path):
+    with open(path, "r") as f:
+        return json.loads(f.read())
+
+
+def write_json(text, path):
+    with open(path, "w") as f:
+        f.write(json.dumps(text))
+
+
+def write_model(model_path, input_base_path, model_size):
+    assert model_size in INTERMEDIATE_SIZE_MAP
+    os.makedirs(model_path, exist_ok=True)
+
+    params = read_json(os.path.join(input_base_path, "params.json"))
+    num_shards = NUM_SHARDS[model_size]
+    n_layers = params["n_layers"]
+    n_heads = params["n_heads"]
+    n_heads_per_shard = n_heads // num_shards
+    dim = params["dim"]
+    dims_per_head = dim // n_heads
+
+    # Load weights
+    if model_size == "7B":
+        # Not shared
+        # (The sharded implementation would also work, but this is simpler.)
+        loaded = torch.load(os.path.join(input_base_path, "consolidated.00.pth"), map_location="cpu")
+    else:
+        # Sharded
+        loaded = [
+            torch.load(os.path.join(input_base_path, f"consolidated.{i:02d}.pth"), map_location="cpu")
+            for i in range(num_shards)
+        ]
+    param_count = 0
+    index_dict = {"weight_map": {}}
+    for layer_i in range(n_layers):
+        filename = "pytorch_model-{:05d}-of-{:05d}.bin".format(
+            layer_i,
+            n_layers + 1,
+        )
+        if model_size == "7B":
+            # Unsharded
+            state_dict = {
+                f"model.decoder.layers.{layer_i}.self_attn.q_proj.weight": loaded[
+                    f"layers.{layer_i}.attention.wq.weight"
+                ],
+                f"model.decoder.layers.{layer_i}.self_attn.k_proj.weight": loaded[
+                    f"layers.{layer_i}.attention.wk.weight"
+                ],
+                f"model.decoder.layers.{layer_i}.self_attn.v_proj.weight": loaded[
+                    f"layers.{layer_i}.attention.wv.weight"
+                ],
+                f"model.decoder.layers.{layer_i}.self_attn.o_proj.weight": loaded[
+                    f"layers.{layer_i}.attention.wo.weight"
+                ],
+                f"model.decoder.layers.{layer_i}.feed_forward.w1.weight": loaded[
+                    f"layers.{layer_i}.feed_forward.w1.weight"
+                ],
+                f"model.decoder.layers.{layer_i}.feed_forward.w2.weight": loaded[
+                    f"layers.{layer_i}.feed_forward.w2.weight"
+                ],
+                f"model.decoder.layers.{layer_i}.feed_forward.w3.weight": loaded[
+                    f"layers.{layer_i}.feed_forward.w3.weight"
+                ],
+                f"model.decoder.layers.{layer_i}.attention_norm.weight": loaded[
+                    f"layers.{layer_i}.attention_norm.weight"
+                ],
+                f"model.decoder.layers.{layer_i}.ffn_norm.weight": loaded[f"layers.{layer_i}.ffn_norm.weight"],
+            }
+        else:
+            # Sharded
+            state_dict = {
+                f"model.decoder.layers.{layer_i}.attention_norm.weight": loaded[0][
+                    f"layers.{layer_i}.attention_norm.weight"
+                ],
+                f"model.decoder.layers.{layer_i}.ffn_norm.weight": loaded[0][f"layers.{layer_i}.ffn_norm.weight"],
+            }
+            state_dict[f"model.decoder.layers.{layer_i}.self_attn.q_proj.weight"] = torch.cat(
+                [
+                    loaded[i][f"layers.{layer_i}.attention.wq.weight"].view(n_heads_per_shard, dims_per_head, dim)
+                    for i in range(num_shards)
+                ],
+                dim=0,
+            ).reshape(dim, dim)
+            state_dict[f"model.decoder.layers.{layer_i}.self_attn.k_proj.weight"] = torch.cat(
+                [
+                    loaded[i][f"layers.{layer_i}.attention.wk.weight"].view(n_heads_per_shard, dims_per_head, dim)
+                    for i in range(num_shards)
+                ],
+                dim=0,
+            ).reshape(dim, dim)
+            state_dict[f"model.decoder.layers.{layer_i}.self_attn.v_proj.weight"] = torch.cat(
+                [
+                    loaded[i][f"layers.{layer_i}.attention.wv.weight"].view(n_heads_per_shard, dims_per_head, dim)
+                    for i in range(num_shards)
+                ],
+                dim=0,
+            ).reshape(dim, dim)
+
+            state_dict[f"model.decoder.layers.{layer_i}.self_attn.o_proj.weight"] = torch.cat(
+                [loaded[i][f"layers.{layer_i}.attention.wo.weight"] for i in range(num_shards)], dim=1
+            )
+            state_dict[f"model.decoder.layers.{layer_i}.feed_forward.w1.weight"] = torch.cat(
+                [loaded[i][f"layers.{layer_i}.feed_forward.w1.weight"] for i in range(num_shards)], dim=0
+            )
+            state_dict[f"model.decoder.layers.{layer_i}.feed_forward.w2.weight"] = torch.cat(
+                [loaded[i][f"layers.{layer_i}.feed_forward.w2.weight"] for i in range(num_shards)], dim=1
+            )
+            state_dict[f"model.decoder.layers.{layer_i}.feed_forward.w3.weight"] = torch.cat(
+                [loaded[i][f"layers.{layer_i}.feed_forward.w3.weight"] for i in range(num_shards)], dim=0
+            )
+
+        for k, v in state_dict.items():
+            index_dict["weight_map"][k] = filename
+            param_count += v.numel()
+        torch.save(state_dict, os.path.join(model_path, filename))
+
+    filename = "pytorch_model-{:05d}-of-{:05d}.bin".format(
+        n_layers,
+        n_layers + 1,
+    )
+    if model_size == "7B":
+        # Unsharded
+        state_dict = {
+            "model.decoder.embed_tokens.weight": loaded["tok_embeddings.weight"],
+            "model.decoder.norm.weight": loaded["norm.weight"],
+            "lm_head.weight": loaded["output.weight"],
+        }
+    else:
+        state_dict = {
+            "model.decoder.norm.weight": loaded[0]["norm.weight"],
+            "model.decoder.embed_tokens.weight": torch.cat(
+                [loaded[i]["tok_embeddings.weight"] for i in range(num_shards)], dim=1
+            ),
+            "lm_head.weight": torch.cat([loaded[i]["output.weight"] for i in range(num_shards)], dim=0),
+        }
+
+    for k, v in state_dict.items():
+        index_dict["weight_map"][k] = filename
+        param_count += v.numel()
+    torch.save(state_dict, os.path.join(model_path, filename))
+
+    # Write configs
+    index_dict["metadata"] = {"total_size": param_count * 2}
+    write_json(index_dict, os.path.join(model_path, "pytorch_model.bin.index.json"))
+    config_out = {
+        "architectures": ["LLaMAForCausalLM"],
+        "bos_token_id": 0,
+        "eos_token_id": 1,
+        "hidden_act": "silu",
+        "hidden_size": params["dim"],
+        "intermediate_size": INTERMEDIATE_SIZE_MAP[model_size],
+        "initializer_range": 0.02,
+        "max_sequence_length": 2048,
+        "model_type": "llama",
+        "num_attention_heads": params["n_heads"],
+        "num_hidden_layers": params["n_layers"],
+        "pad_token_id": -1,
+        "rms_norm_eps": params["norm_eps"],
+        "torch_dtype": "float16",
+        "transformers_version": "4.27.0.dev0",
+        "use_cache": True,
+        "vocab_size": 32000,
+    }
+    write_json(
+        config_out,
+        os.path.join(model_path, "config.json"),
+    )
+    generation_config = {
+        "_from_model_config": True,
+        "bos_token_id": 0,
+        "eos_token_id": 1,
+        "pad_token_id": -1,
+        "transformers_version": "4.27.0.dev0",
+    }
+    write_json(
+        generation_config,
+        os.path.join(model_path, "generation_config.json"),
+    )
+
+
+def write_tokenizer(tokenizer_path, input_tokenizer_path):
+    os.makedirs(tokenizer_path, exist_ok=True)
+    write_json({}, os.path.join(tokenizer_path, "special_tokens_map.json"))
+    write_json(
+        {
+            "bos_token": "",
+            "eos_token": "",
+            "model_max_length": int(1e30),
+            "tokenizer_class": "LLaMATokenizer",
+            "unk_token": "",
+        },
+        os.path.join(tokenizer_path, "tokenizer_config.json"),
+    )
+    shutil.copyfile(input_tokenizer_path, os.path.join(tokenizer_path, "tokenizer.model"))
+
+
+def main():
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "--input_dir",
+        help="Location of LLaMA weights, which contains tokenizer.model and model folders",
+    )
+    parser.add_argument(
+        "--model_size",
+        choices=["7B", "13B", "30B", "65B"],
+    )
+    parser.add_argument(
+        "--output_dir",
+        help="Location to write HF model and tokenizer",
+    )
+    args = parser.parse_args()
+    write_model(
+        model_path=os.path.join(args.output_dir, "llama-{}".format(args.model_size).lower()),
+        input_base_path=os.path.join(args.input_dir, args.model_size),
+        model_size=args.model_size,
+    )
+    write_tokenizer(
+        tokenizer_path=os.path.join(args.output_dir, "tokenizer"),
+        input_tokenizer_path=os.path.join(args.input_dir, "tokenizer.model"),
+    )
+
+
+if __name__ == "__main__":
+    main()

llama/modeling_llama.py

@@ -0,0 +1,941 @@
+# coding=utf-8
+# Copyright 2022 The Fairseq Authors and The 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 LLaMA model."""
+import math
+from typing import List, Optional, Tuple, Union
+
+import torch
+import torch.utils.checkpoint
+from torch import nn
+from torch.nn import CrossEntropyLoss
+
+from transformers.activations import ACT2FN
+from transformers.modeling_outputs import (
+    BaseModelOutputWithPast,
+    CausalLMOutputWithPast,
+)
+from transformers.modeling_utils import PreTrainedModel
+from transformers.utils import (
+    add_code_sample_docstrings,
+    add_start_docstrings,
+    add_start_docstrings_to_model_forward,
+    logging,
+    replace_return_docstrings,
+)
+from .configuration_llama import LLaMAConfig
+
+
+logger = logging.get_logger(__name__)
+
+_CHECKPOINT_FOR_DOC = "llama-7b"
+_CONFIG_FOR_DOC = "LLaMAConfig"
+
+
+def _make_causal_mask(input_ids_shape: torch.Size, dtype: torch.dtype, past_key_values_length: int = 0):
+    """
+    Make causal mask used for bi-directional self-attention.
+    """
+    bsz, tgt_len = input_ids_shape
+    mask = torch.full((tgt_len, tgt_len), torch.tensor(torch.finfo(dtype).min))
+    mask_cond = torch.arange(mask.size(-1))
+    mask.masked_fill_(mask_cond < (mask_cond + 1).view(mask.size(-1), 1), 0)
+    mask = mask.to(dtype)
+
+    if past_key_values_length > 0:
+        mask = torch.cat([torch.zeros(tgt_len, past_key_values_length, dtype=dtype), mask], dim=-1)
+    return mask[None, None, :, :].expand(bsz, 1, tgt_len, tgt_len + past_key_values_length)
+
+
+def _expand_mask(mask: torch.Tensor, dtype: torch.dtype, tgt_len: Optional[int] = None):
+    """
+    Expands attention_mask from `[bsz, seq_len]` to `[bsz, 1, tgt_seq_len, src_seq_len]`.
+    """
+    bsz, src_len = mask.size()
+    tgt_len = tgt_len if tgt_len is not None else src_len
+
+    expanded_mask = mask[:, None, None, :].expand(bsz, 1, tgt_len, src_len).to(dtype)
+
+    inverted_mask = 1.0 - expanded_mask
+
+    return inverted_mask.masked_fill(inverted_mask.to(torch.bool), torch.finfo(dtype).min)
+
+
+class RMSNorm(torch.nn.Module):
+    def __init__(self, dim: int, eps: float = 1e-6):
+        super().__init__()
+        self.eps = eps
+        self.weight = nn.Parameter(torch.ones(dim))
+
+    def _norm(self, x):
+        return x * torch.rsqrt(x.pow(2).mean(-1, keepdim=True) + self.eps)
+
+    def forward(self, x):
+        output = self._norm(x.float()).type_as(x)
+        return output * self.weight
+
+
+class LLaMAFeedForward(nn.Module):
+    def __init__(
+        self,
+        hidden_size: int,
+        intermediate_size: int,
+        hidden_act: str,
+    ):
+        super().__init__()
+        self.w1 = nn.Linear(hidden_size, intermediate_size, bias=False)
+        self.w2 = nn.Linear(intermediate_size, hidden_size, bias=False)
+        self.w3 = nn.Linear(hidden_size, intermediate_size, bias=False)
+        self.act_fn = ACT2FN[hidden_act]
+
+    def forward(self, x):
+        return self.w2(self.act_fn(self.w1(x)) * self.w3(x))
+
+
+class LLaMAAttention(nn.Module):
+    """Multi-headed attention from 'Attention Is All You Need' paper"""
+
+    def __init__(
+        self,
+        hidden_size: int,
+        num_heads: int,
+        complex_frequencies: torch.Tensor,
+    ):
+        super().__init__()
+        self.hidden_size = hidden_size
+        self.num_heads = num_heads
+        self.head_dim = hidden_size // num_heads
+
+        if (self.head_dim * num_heads) != self.hidden_size:
+            raise ValueError(
+                f"hidden_size must be divisible by num_heads (got `hidden_size`: {self.hidden_size}"
+                f" and `num_heads`: {num_heads})."
+            )
+        self.q_proj = nn.Linear(
+            hidden_size,
+            num_heads * self.head_dim,
+            bias=False,
+        )
+        self.k_proj = nn.Linear(
+            hidden_size,
+            num_heads * self.head_dim,
+            bias=False,
+        )
+        self.v_proj = nn.Linear(
+            hidden_size,
+            num_heads * self.head_dim,
+            bias=False,
+        )
+        self.o_proj = nn.Linear(
+            num_heads * self.head_dim,
+            hidden_size,
+            bias=False,
+        )
+        self.complex_frequencies = complex_frequencies
+
+    def _shape(self, tensor: torch.Tensor, seq_len: int, bsz: int):
+        return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        past_key_value: Optional[Tuple[torch.Tensor]] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        layer_head_mask: Optional[torch.Tensor] = None,
+        output_attentions: bool = False,
+    ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
+        """Input shape: Batch x Time x Channel"""
+
+        self.complex_frequencies = self.complex_frequencies.to(hidden_states.device)
+
+        bsz, tgt_len, _ = hidden_states.size()
+
+        # get query proj
+        query_states = self.q_proj(hidden_states).view(bsz, tgt_len, self.num_heads, self.head_dim)
+        key_states = self.k_proj(hidden_states).view(bsz, tgt_len, self.num_heads, self.head_dim)
+        value_states = self.v_proj(hidden_states).view(bsz, tgt_len, self.num_heads, self.head_dim)
+
+        if past_key_value is not None:
+            start = past_key_value[0].shape[2]
+        else:
+            start = 0
+
+        sliced_complex_frequencies = self.complex_frequencies[start : start + tgt_len]
+        query_states, key_states = apply_rotary_emb(
+            query_states=query_states, key_states=key_states, complex_frequencies=sliced_complex_frequencies
+        )
+
+        # get key, value proj
+        key_states = self._shape(key_states, -1, bsz)
+        value_states = self._shape(value_states, -1, bsz)
+        if past_key_value is not None:
+            # reuse k, v, self_attention
+            key_states = torch.cat([past_key_value[0], key_states], dim=2)
+            value_states = torch.cat([past_key_value[1], value_states], dim=2)
+
+        past_key_value = (key_states, value_states)
+        proj_shape = (bsz * self.num_heads, -1, self.head_dim)
+        query_states = self._shape(query_states, tgt_len, bsz).view(*proj_shape)
+        key_states = key_states.view(*proj_shape)
+        value_states = value_states.view(*proj_shape)
+
+        src_len = key_states.size(1)
+        attn_weights = torch.bmm(query_states, key_states.transpose(1, 2)) / math.sqrt(self.head_dim)
+
+        if attn_weights.size() != (bsz * self.num_heads, tgt_len, src_len):
+            raise ValueError(
+                f"Attention weights should be of size {(bsz * self.num_heads, tgt_len, src_len)}, but is"
+                f" {attn_weights.size()}"
+            )
+
+        if attention_mask is not None:
+            if attention_mask.size() != (bsz, 1, tgt_len, src_len):
+                raise ValueError(
+                    f"Attention mask should be of size {(bsz, 1, tgt_len, src_len)}, but is {attention_mask.size()}"
+                )
+            attn_weights = attn_weights.view(bsz, self.num_heads, tgt_len, src_len) + attention_mask
+            attn_weights = torch.max(attn_weights, torch.tensor(torch.finfo(attn_weights.dtype).min))
+            attn_weights = attn_weights.view(bsz * self.num_heads, tgt_len, src_len)
+
+        # upcast to fp32 if the weights are in fp16. Please see https://github.com/huggingface/transformers/pull/17437
+        if attn_weights.dtype == torch.float16:
+            attn_weights = nn.functional.softmax(attn_weights, dim=-1, dtype=torch.float32).to(torch.float16)
+        else:
+            attn_weights = nn.functional.softmax(attn_weights, dim=-1)
+
+        if layer_head_mask is not None:
+            if layer_head_mask.size() != (self.num_heads,):
+                raise ValueError(
+                    f"Head mask for a single layer should be of size {(self.num_heads,)}, but is"
+                    f" {layer_head_mask.size()}"
+                )
+            attn_weights = layer_head_mask.view(1, -1, 1, 1) * attn_weights.view(bsz, self.num_heads, tgt_len, src_len)
+            attn_weights = attn_weights.view(bsz * self.num_heads, tgt_len, src_len)
+
+        if output_attentions:
+            # this operation is a bit awkward, but it's required to
+            # make sure that attn_weights keeps its gradient.
+            # In order to do so, attn_weights have to be reshaped
+            # twice and have to be reused in the following
+            attn_weights_reshaped = attn_weights.view(bsz, self.num_heads, tgt_len, src_len)
+            attn_weights = attn_weights_reshaped.view(bsz * self.num_heads, tgt_len, src_len)
+        else:
+            attn_weights_reshaped = None
+
+        attn_output = torch.bmm(attn_weights, value_states)
+
+        if attn_output.size() != (bsz * self.num_heads, tgt_len, self.head_dim):
+            raise ValueError(
+                f"`attn_output` should be of size {(bsz, self.num_heads, tgt_len, self.head_dim)}, but is"
+                f" {attn_output.size()}"
+            )
+
+        attn_output = attn_output.view(bsz, self.num_heads, tgt_len, self.head_dim)
+        attn_output = attn_output.transpose(1, 2)
+
+        attn_output = attn_output.reshape(bsz, tgt_len, self.hidden_size)
+
+        attn_output = self.o_proj(attn_output)
+
+        return attn_output, attn_weights_reshaped, past_key_value
+
+
+class LLaMADecoderLayer(nn.Module):
+    def __init__(self, config: LLaMAConfig):
+        super().__init__()
+        self.hidden_size = config.hidden_size
+        complex_frequencies = precompute_complex_frequencies(
+            head_dim=self.hidden_size // config.num_attention_heads,
+            length=config.max_sequence_length * 2,
+        )
+        self.self_attn = LLaMAAttention(
+            hidden_size=self.hidden_size,
+            num_heads=config.num_attention_heads,
+            complex_frequencies=complex_frequencies,
+        )
+        self.feed_forward = LLaMAFeedForward(
+            hidden_size=self.hidden_size,
+            intermediate_size=config.intermediate_size,
+            hidden_act=config.hidden_act,
+        )
+        self.attention_norm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+        self.ffn_norm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        attention_mask: Optional[torch.Tensor] = None,
+        layer_head_mask: Optional[torch.Tensor] = None,
+        output_attentions: Optional[bool] = False,
+        use_cache: Optional[bool] = False,
+        past_key_value: Optional[Tuple[torch.Tensor]] = None,
+    ) -> Tuple[torch.FloatTensor, Optional[Tuple[torch.FloatTensor, torch.FloatTensor]]]:
+        """
+        Args:
+            hidden_states (`torch.FloatTensor`): input to the layer of shape `(batch, seq_len, embed_dim)`
+            attention_mask (`torch.FloatTensor`, *optional*): attention mask of size
+                `(batch, 1, tgt_len, src_len)` where padding elements are indicated by very large negative values.
+            layer_head_mask (`torch.FloatTensor`, *optional*): mask for attention heads in a given layer of size
+                `(encoder_attention_heads,)`.
+            output_attentions (`bool`, *optional*):
+                Whether or not to return the attentions tensors of all attention layers. See `attentions` under
+                returned tensors for more detail.
+            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`).
+            past_key_value (`Tuple(torch.FloatTensor)`, *optional*): cached past key and value projection states
+        """
+
+        residual = hidden_states
+
+        hidden_states = self.attention_norm(hidden_states)
+
+        # Self Attention
+        hidden_states, self_attn_weights, present_key_value = self.self_attn(
+            hidden_states=hidden_states,
+            past_key_value=past_key_value,
+            attention_mask=attention_mask,
+            layer_head_mask=layer_head_mask,
+            output_attentions=output_attentions,
+        )
+        hidden_states = residual + hidden_states
+
+        # Fully Connected
+        residual = hidden_states
+        hidden_states = self.ffn_norm(hidden_states)
+        hidden_states = self.feed_forward(hidden_states)
+        hidden_states = residual + hidden_states
+
+        outputs = (hidden_states,)
+
+        if output_attentions:
+            outputs += (self_attn_weights,)
+
+        if use_cache:
+            outputs += (present_key_value,)
+
+        return outputs
+
+
+LLAMA_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, pruning heads
+    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 ([`LLaMAConfig`]):
+            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.
+"""
+
+
+@add_start_docstrings(
+    "The bare OPT Model outputting raw hidden-states without any specific head on top.",
+    LLAMA_START_DOCSTRING,
+)
+class LLaMAPreTrainedModel(PreTrainedModel):
+    config_class = LLaMAConfig
+    base_model_prefix = "model"
+    supports_gradient_checkpointing = True
+    _no_split_modules = ["LLaMADecoderLayer"]
+    _keys_to_ignore_on_load_unexpected = [r"decoder\.version"]
+
+    def _init_weights(self, module):
+        std = self.config.initializer_range
+        if isinstance(module, nn.Linear):
+            module.weight.data.normal_(mean=0.0, std=std)
+            if module.bias is not None:
+                module.bias.data.zero_()
+        elif isinstance(module, nn.Embedding):
+            module.weight.data.normal_(mean=0.0, std=std)
+            if module.padding_idx is not None:
+                module.weight.data[module.padding_idx].zero_()
+
+    def _set_gradient_checkpointing(self, module, value=False):
+        if isinstance(module, (LLaMADecoder)):
+            module.gradient_checkpointing = value
+
+
+LLAMA_INPUTS_DOCSTRING = r"""
+    Args:
+        input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`):
+            Indices of input sequence tokens in the vocabulary. Padding will be ignored by default should you provide
+            it.
+
+            Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and
+            [`PreTrainedTokenizer.__call__`] for details.
+
+            [What are input IDs?](../glossary#input-ids)
+        attention_mask (`torch.Tensor` 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)
+
+            Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and
+            [`PreTrainedTokenizer.__call__`] for details.
+
+            If `past_key_values` is used, optionally only the last `decoder_input_ids` have to be input (see
+            `past_key_values`).
+
+            If you want to change padding behavior, you should read [`modeling_opt._prepare_decoder_attention_mask`]
+            and modify to your needs. See diagram 1 in [the paper](https://arxiv.org/abs/1910.13461) for more
+            information on the default strategy.
+        head_mask (`torch.Tensor` of shape `(encoder_layers, encoder_attention_heads)`, *optional*):
+            Mask to nullify selected heads of the attention modules in the encoder. Mask values selected in `[0, 1]`:
+
+            - 1 indicates the head is **not masked**,
+            - 0 indicates the head is **masked**.
+
+        past_key_values (`tuple(tuple(torch.FloatTensor))`, *optional*, returned when `use_cache=True` is passed or when `config.use_cache=True`):
+            Tuple of `tuple(torch.FloatTensor)` of length `config.n_layers`, with each tuple having 2 tensors of shape
+            `(batch_size, num_heads, sequence_length, embed_size_per_head)`) and 2 additional tensors of shape
+            `(batch_size, num_heads, encoder_sequence_length, embed_size_per_head)`.
+
+            Contains pre-computed hidden-states (key and values in the self-attention blocks and in the cross-attention
+            blocks) that can be used (see `past_key_values` input) to speed up sequential decoding.
+
+            If `past_key_values` are used, the user can optionally input only the last `decoder_input_ids` (those that
+            don't have their past key value states given to this model) of shape `(batch_size, 1)` instead of all
+            `decoder_input_ids` of shape `(batch_size, sequence_length)`.
+        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.
+        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 [`~utils.ModelOutput`] instead of a plain tuple.
+"""
+
+
+class LLaMADecoder(LLaMAPreTrainedModel):
+    """
+    Transformer decoder consisting of *config.num_hidden_layers* layers. Each layer is a [`LLaMADecoderLayer`]
+
+    Args:
+        config: LLaMAConfig
+    """
+
+    def __init__(self, config: LLaMAConfig):
+        super().__init__(config)
+        self.padding_idx = config.pad_token_id
+
+        self.vocab_size = config.vocab_size
+
+        self.embed_tokens = nn.Embedding(config.vocab_size, config.hidden_size, self.padding_idx)
+
+        self.layers = nn.ModuleList([LLaMADecoderLayer(config) for _ in range(config.num_hidden_layers)])
+        self.norm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+
+        self.gradient_checkpointing = False
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def get_input_embeddings(self):
+        return self.embed_tokens
+
+    def set_input_embeddings(self, value):
+        self.embed_tokens = value
+
+    # Copied from transformers.models.bart.modeling_bart.BartDecoder._prepare_decoder_attention_mask
+    def _prepare_decoder_attention_mask(self, attention_mask, input_shape, inputs_embeds, past_key_values_length):
+        # create causal mask
+        # [bsz, seq_len] -> [bsz, 1, tgt_seq_len, src_seq_len]
+        combined_attention_mask = None
+        if input_shape[-1] > 1:
+            combined_attention_mask = _make_causal_mask(
+                input_shape, inputs_embeds.dtype, past_key_values_length=past_key_values_length
+            ).to(inputs_embeds.device)
+
+        if attention_mask is not None:
+            # [bsz, seq_len] -> [bsz, 1, tgt_seq_len, src_seq_len]
+            expanded_attn_mask = _expand_mask(attention_mask, inputs_embeds.dtype, tgt_len=input_shape[-1]).to(
+                inputs_embeds.device
+            )
+            combined_attention_mask = (
+                expanded_attn_mask if combined_attention_mask is None else expanded_attn_mask + combined_attention_mask
+            )
+
+        return combined_attention_mask
+
+    def forward(
+        self,
+        input_ids: torch.LongTensor = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        head_mask: Optional[torch.Tensor] = None,
+        past_key_values: Optional[List[torch.FloatTensor]] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+    ) -> Union[Tuple, BaseModelOutputWithPast]:
+        r"""
+        Args:
+            input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`):
+                Indices of input sequence tokens in the vocabulary. Padding will be ignored by default should you
+                provide it.
+
+                Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and
+                [`PreTrainedTokenizer.__call__`] for details.
+
+                [What are input IDs?](../glossary#input-ids)
+            attention_mask (`torch.Tensor` 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.Tensor` of shape `(num_hidden_layers, num_attention_heads)`, *optional*):
+                Mask to nullify selected heads of the attention modules. Mask values selected in `[0, 1]`:
+
+                - 1 indicates the head is **not masked**,
+                - 0 indicates the head is **masked**.
+
+            past_key_values (`tuple(tuple(torch.FloatTensor))`, *optional*, returned when `use_cache=True` is passed or when `config.use_cache=True`):
+                Tuple of `tuple(torch.FloatTensor)` of length `config.n_layers`, with each tuple having 2 tensors of
+                shape `(batch_size, num_heads, sequence_length, embed_size_per_head)`) and 2 additional tensors of
+
+                Contains pre-computed hidden-states (key and values in the self-attention blocks and in the
+                cross-attention blocks) that can be used (see `past_key_values` input) to speed up sequential decoding.
+
+                If `past_key_values` are used, the user can optionally input only the last `decoder_input_ids` (those
+                that don't have their past key value states given to this model) of shape `(batch_size, 1)` instead of
+                all `decoder_input_ids` of shape `(batch_size, sequence_length)`.
+
+            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.
+            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 [`~utils.ModelOutput`] instead of a plain tuple.
+        """
+        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
+        )
+        use_cache = use_cache if use_cache is not None else self.config.use_cache
+
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        # retrieve input_ids and inputs_embeds
+        if input_ids is not None and inputs_embeds is not None:
+            raise ValueError("You cannot specify both decoder_input_ids and decoder_inputs_embeds at the same time")
+        elif input_ids is not None:
+            input_shape = input_ids.size()
+            input_ids = input_ids.view(-1, input_shape[-1])
+        elif inputs_embeds is not None:
+            input_shape = inputs_embeds.size()[:-1]
+        else:
+            raise ValueError("You have to specify either decoder_input_ids or decoder_inputs_embeds")
+
+        past_key_values_length = past_key_values[0][0].shape[2] if past_key_values is not None else 0
+
+        if inputs_embeds is None:
+            inputs_embeds = self.embed_tokens(input_ids)
+
+        # embed positions
+        if attention_mask is None:
+            attention_mask = torch.ones(inputs_embeds.shape[:2], dtype=torch.bool, device=inputs_embeds.device)
+
+        attention_mask = self._prepare_decoder_attention_mask(
+            attention_mask, input_shape, inputs_embeds, past_key_values_length
+        )
+
+        hidden_states = inputs_embeds
+
+        if self.gradient_checkpointing and self.training:
+            if use_cache:
+                logger.warning_once(
+                    "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
+                )
+                use_cache = False
+
+        # decoder layers
+        all_hidden_states = () if output_hidden_states else None
+        all_self_attns = () if output_attentions else None
+        next_decoder_cache = () if use_cache else None
+
+        # check if head_mask has a correct number of layers specified if desired
+        for attn_mask, mask_name in zip([head_mask], ["head_mask"]):
+            if attn_mask is not None:
+                if attn_mask.size()[0] != (len(self.layers)):
+                    raise ValueError(
+                        f"The `{mask_name}` should be specified for {len(self.layers)} layers, but it is for"
+                        f" {head_mask.size()[0]}."
+                    )
+
+        for idx, decoder_layer in enumerate(self.layers):
+            if output_hidden_states:
+                all_hidden_states += (hidden_states,)
+
+            past_key_value = past_key_values[idx] if past_key_values is not None else None
+
+            if self.gradient_checkpointing and self.training:
+
+                def create_custom_forward(module):
+                    def custom_forward(*inputs):
+                        # None for past_key_value
+                        return module(*inputs, output_attentions, None)
+
+                    return custom_forward
+
+                layer_outputs = torch.utils.checkpoint.checkpoint(
+                    create_custom_forward(decoder_layer),
+                    hidden_states,
+                    attention_mask,
+                    head_mask[idx] if head_mask is not None else None,
+                    None,
+                )
+            else:
+                layer_outputs = decoder_layer(
+                    hidden_states,
+                    attention_mask=attention_mask,
+                    layer_head_mask=(head_mask[idx] if head_mask is not None else None),
+                    past_key_value=past_key_value,
+                    output_attentions=output_attentions,
+                    use_cache=use_cache,
+                )
+
+            hidden_states = layer_outputs[0]
+
+            if use_cache:
+                next_decoder_cache += (layer_outputs[2 if output_attentions else 1],)
+
+            if output_attentions:
+                all_self_attns += (layer_outputs[1],)
+
+        hidden_states = self.norm(hidden_states)
+
+        # add hidden states from the last decoder layer
+        if output_hidden_states:
+            all_hidden_states += (hidden_states,)
+
+        next_cache = next_decoder_cache if use_cache else None
+        if not return_dict:
+            return tuple(v for v in [hidden_states, next_cache, all_hidden_states, all_self_attns] if v is not None)
+        return BaseModelOutputWithPast(
+            last_hidden_state=hidden_states,
+            past_key_values=next_cache,
+            hidden_states=all_hidden_states,
+            attentions=all_self_attns,
+        )
+
+
+@add_start_docstrings(
+    "The bare LLaMA Model outputting raw hidden-states without any specific head on top.",
+    LLAMA_START_DOCSTRING,
+)
+class LLaMAModel(LLaMAPreTrainedModel):
+    def __init__(self, config: LLaMAConfig):
+        super().__init__(config)
+        self.decoder = LLaMADecoder(config)
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def get_input_embeddings(self):
+        return self.decoder.embed_tokens
+
+    def set_input_embeddings(self, value):
+        self.decoder.embed_tokens = value
+
+    def get_decoder(self):
+        return self.decoder
+
+    @add_start_docstrings_to_model_forward(LLAMA_INPUTS_DOCSTRING)
+    @add_code_sample_docstrings(
+        checkpoint=_CHECKPOINT_FOR_DOC,
+        output_type=BaseModelOutputWithPast,
+        config_class=_CONFIG_FOR_DOC,
+    )
+    def forward(
+        self,
+        input_ids: torch.LongTensor = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        head_mask: Optional[torch.Tensor] = None,
+        past_key_values: Optional[List[torch.FloatTensor]] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+    ) -> Union[Tuple, BaseModelOutputWithPast]:
+        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
+        )
+        use_cache = use_cache if use_cache is not None else self.config.use_cache
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        # decoder outputs consists of (dec_features, past_key_value, dec_hidden, dec_attn)
+        decoder_outputs = self.decoder(
+            input_ids=input_ids,
+            attention_mask=attention_mask,
+            head_mask=head_mask,
+            past_key_values=past_key_values,
+            inputs_embeds=inputs_embeds,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+
+        if not return_dict:
+            return decoder_outputs
+
+        return BaseModelOutputWithPast(
+            last_hidden_state=decoder_outputs.last_hidden_state,
+            past_key_values=decoder_outputs.past_key_values,
+            hidden_states=decoder_outputs.hidden_states,
+            attentions=decoder_outputs.attentions,
+        )
+
+
+class LLaMAForCausalLM(LLaMAPreTrainedModel):
+    _keys_to_ignore_on_load_missing = [r"lm_head.weight"]
+
+    def __init__(self, config):
+        super().__init__(config)
+        self.model = LLaMAModel(config)
+
+        self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def get_input_embeddings(self):
+        return self.model.decoder.embed_tokens
+
+    def set_input_embeddings(self, value):
+        self.model.decoder.embed_tokens = value
+
+    def get_output_embeddings(self):
+        return self.lm_head
+
+    def set_output_embeddings(self, new_embeddings):
+        self.lm_head = new_embeddings
+
+    def set_decoder(self, decoder):
+        self.model.decoder = decoder
+
+    def get_decoder(self):
+        return self.model.decoder
+
+    @replace_return_docstrings(output_type=CausalLMOutputWithPast, config_class=_CONFIG_FOR_DOC)
+    def forward(
+        self,
+        input_ids: torch.LongTensor = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        head_mask: Optional[torch.Tensor] = None,
+        past_key_values: Optional[List[torch.FloatTensor]] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        labels: Optional[torch.LongTensor] = None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+    ) -> Union[Tuple, CausalLMOutputWithPast]:
+        r"""
+        Args:
+            input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`):
+                Indices of input sequence tokens in the vocabulary. Padding will be ignored by default should you
+                provide it.
+
+                Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and
+                [`PreTrainedTokenizer.__call__`] for details.
+
+                [What are input IDs?](../glossary#input-ids)
+            attention_mask (`torch.Tensor` 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.Tensor` of shape `(num_hidden_layers, num_attention_heads)`, *optional*):
+                Mask to nullify selected heads of the attention modules. Mask values selected in `[0, 1]`:
+
+                - 1 indicates the head is **not masked**,
+                - 0 indicates the head is **masked**.
+
+            past_key_values (`tuple(tuple(torch.FloatTensor))`, *optional*, returned when `use_cache=True` is passed or when `config.use_cache=True`):
+                Tuple of `tuple(torch.FloatTensor)` of length `config.n_layers`, with each tuple having 2 tensors of
+                shape `(batch_size, num_heads, sequence_length, embed_size_per_head)`) and 2 additional tensors of
+                shape `(batch_size, num_heads, encoder_sequence_length, embed_size_per_head)`. The two additional
+                tensors are only required when the model is used as a decoder in a Sequence to Sequence model.
+
+                Contains pre-computed hidden-states (key and values in the self-attention blocks and in the
+                cross-attention blocks) that can be used (see `past_key_values` input) to speed up sequential decoding.
+
+                If `past_key_values` are used, the user can optionally input only the last `decoder_input_ids` (those
+                that don't have their past key value states given to this model) of shape `(batch_size, 1)` instead of
+                all `decoder_input_ids` of shape `(batch_size, sequence_length)`.
+            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.
+            labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+                Labels for computing the masked language modeling loss. Indices should either be in `[0, ...,
+                config.vocab_size]` or -100 (see `input_ids` docstring). Tokens with indices set to `-100` are ignored
+                (masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`.
+            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 [`~utils.ModelOutput`] instead of a plain tuple.
+
+        Returns:
+
+        Example:
+
+        ```python
+        >>> from transformers import AutoTokenizer, LLaMAForCausalLM
+
+        >>> model = LLaMAForCausalLM.from_pretrained(PATH_TO_CONVERTED_WEIGHTS)
+        >>> tokenizer = AutoTokenizer.from_pretrained(PATH_TO_CONVERTED_TOKENIZER)
+
+        >>> prompt = "Hey, are you consciours? Can you talk to me?"
+        >>> inputs = tokenizer(prompt, return_tensors="pt")
+
+        >>> # Generate
+        >>> generate_ids = model.generate(inputs.input_ids, max_length=30)
+        >>> tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
+        "Hey, are you consciours? Can you talk to me?\nI'm not consciours, but I can talk to you."
+        ```"""
+
+        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
+
+        # decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
+        outputs = self.model.decoder(
+            input_ids=input_ids,
+            attention_mask=attention_mask,
+            head_mask=head_mask,
+            past_key_values=past_key_values,
+            inputs_embeds=inputs_embeds,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+
+        logits = self.lm_head(outputs[0]).contiguous()
+
+        loss = None
+        if labels is not None:
+            # Shift so that tokens < n predict n
+            shift_logits = logits[..., :-1, :].contiguous()
+            shift_labels = labels[..., 1:].contiguous()
+            # Flatten the tokens
+            loss_fct = CrossEntropyLoss()
+            loss = loss_fct(shift_logits.view(-1, self.config.vocab_size), shift_labels.view(-1))
+
+        if not return_dict:
+            output = (logits,) + outputs[1:]
+            return (loss,) + output if loss is not None else output
+
+        return CausalLMOutputWithPast(
+            loss=loss,
+            logits=logits,
+            past_key_values=outputs.past_key_values,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+        )
+
+    def prepare_inputs_for_generation(
+        self, input_ids, past_key_values=None, attention_mask=None, inputs_embeds=None, **kwargs
+    ):
+        if past_key_values:
+            input_ids = input_ids[:, -1:]
+
+        # if `inputs_embeds` are passed, we only want to use them in the 1st generation step
+        if inputs_embeds is not None and past_key_values is None:
+            model_inputs = {"inputs_embeds": inputs_embeds}
+        else:
+            model_inputs = {"input_ids": input_ids}
+
+        model_inputs.update(
+            {
+                "past_key_values": past_key_values,
+                "use_cache": kwargs.get("use_cache"),
+                "attention_mask": attention_mask,
+            }
+        )
+        return model_inputs
+
+    @staticmethod
+    def _reorder_cache(past_key_values, beam_idx):
+        reordered_past = ()
+        for layer_past in past_key_values:
+            reordered_past += (tuple(past_state.index_select(0, beam_idx) for past_state in layer_past),)
+        return reordered_past
+
+
+def precompute_complex_frequencies(head_dim: int, length: int, theta: float = 10000.0):
+    frequencies = 1.0 / (theta ** (torch.arange(0, head_dim, 2)[: (head_dim // 2)].float() / head_dim))
+    t = torch.arange(length, device=frequencies.device)
+    frequencies = torch.outer(t, frequencies).float()
+    return torch.polar(torch.ones_like(frequencies), frequencies)  # complex64
+
+
+def apply_rotary_emb(
+    query_states: torch.Tensor,
+    key_states: torch.Tensor,
+    complex_frequencies: torch.Tensor,
+) -> Tuple[torch.Tensor, torch.Tensor]:
+    query_states_complex = torch.view_as_complex(query_states.float().reshape(*key_states.shape[:-1], -1, 2))
+    key_states_complex = torch.view_as_complex(key_states.float().reshape(*key_states.shape[:-1], -1, 2))
+    complex_frequencies = reshape_for_broadcast(complex_frequencies, query_states_complex)
+    output_query_states = torch.view_as_real(query_states_complex * complex_frequencies).flatten(3)
+    output_key_states = torch.view_as_real(key_states_complex * complex_frequencies).flatten(3)
+    return output_query_states.type_as(query_states), output_key_states.type_as(key_states)
+
+
+def reshape_for_broadcast(complex_frequencies: torch.Tensor, x: torch.Tensor):
+    ndim = x.ndim
+    assert 0 <= 1 < ndim
+    assert complex_frequencies.shape == (x.shape[1], x.shape[-1])
+    shape = [d if i == 1 or i == ndim - 1 else 1 for i, d in enumerate(x.shape)]
+    return complex_frequencies.view(*shape)

llama/tokenization_llama.py

@@ -0,0 +1,209 @@
+# coding=utf-8
+# Copyright 2022 The FAIR team of Meta AI and The 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.
+"""Tokenization classes for LLaMA."""
+import os
+import re
+from shutil import copyfile
+from typing import Any, Dict, List, Optional, Tuple
+
+import sentencepiece as spm
+
+from transformers.tokenization_utils import PreTrainedTokenizer
+from transformers.utils import logging
+
+
+logger = logging.get_logger(__name__)
+
+VOCAB_FILES_NAMES = {"vocab_file": "tokenizer.model"}
+
+PRETRAINED_VOCAB_FILES_MAP = {}
+
+
+class LLaMATokenizer(PreTrainedTokenizer):
+    """
+    Construct a LLaMA tokenizer. Based on byte-level Byte-Pair-Encoding.
+
+    Args:
+        vocab_file (`str`):
+            Path to the vocabulary file.
+    """
+
+    vocab_files_names = VOCAB_FILES_NAMES
+    pretrained_vocab_files_map = PRETRAINED_VOCAB_FILES_MAP
+    model_input_names = ["input_ids", "attention_mask"]
+
+    def __init__(
+        self,
+        vocab_file,
+        unk_token="",
+        bos_token="",
+        eos_token="",
+        sp_model_kwargs: Optional[Dict[str, Any]] = None,
+        add_bos_token=False,
+        add_eos_token=False,
+        **kwargs,
+    ):
+        self.sp_model_kwargs = {} if sp_model_kwargs is None else sp_model_kwargs
+        super().__init__(bos_token=bos_token, eos_token=eos_token, unk_token=unk_token, **kwargs)
+        self.vocab_file = vocab_file
+        self.add_bos_token = add_bos_token
+        self.add_eos_token = add_eos_token
+        self.sp_model = spm.SentencePieceProcessor(**self.sp_model_kwargs)
+        self.sp_model.Load(vocab_file)
+
+        """ Initialisation"""
+
+    @property
+    def vocab_size(self):
+        """Returns vocab size"""
+        return self.sp_model.get_piece_size()
+
+    @property
+    def bos_token_id(self) -> Optional[int]:
+        return self.sp_model.bos_id()
+
+    @property
+    def eos_token_id(self) -> Optional[int]:
+        return self.sp_model.eos_id()
+
+    def get_vocab(self):
+        """Returns vocab as a dict"""
+        vocab = {self.convert_ids_to_tokens(i): i for i in range(self.vocab_size)}
+        vocab.update(self.added_tokens_encoder)
+        return vocab
+
+    def _tokenize(self, text):
+        """Returns a tokenized string."""
+        return self.sp_model.encode(text, out_type=str)
+
+    def _convert_token_to_id(self, token):
+        """Converts a token (str) in an id using the vocab."""
+        return self.sp_model.piece_to_id(token)
+
+    def _convert_id_to_token(self, index):
+        """Converts an index (integer) in a token (str) using the vocab."""
+        token = self.sp_model.IdToPiece(index)
+        return token
+
+    def convert_tokens_to_string(self, tokens):
+        """Converts a sequence of tokens (string) in a single string."""
+        current_sub_tokens = []
+        out_string = ""
+        prev_is_special = False
+        for token in tokens:
+            # make sure that special tokens are not decoded using sentencepiece model
+            if token in self.all_special_tokens:
+                if not prev_is_special:
+                    out_string += " "
+                out_string += self.sp_model.decode(current_sub_tokens) + token
+                prev_is_special = True
+                current_sub_tokens = []
+            else:
+                current_sub_tokens.append(token)
+                prev_is_special = False
+        out_string += self.sp_model.decode(current_sub_tokens)
+        return out_string.strip()
+
+    def save_vocabulary(self, save_directory, filename_prefix: Optional[str] = None) -> Tuple[str]:
+        """
+        Save the vocabulary and special tokens file to a directory.
+
+        Args:
+            save_directory (`str`):
+                The directory in which to save the vocabulary.
+
+        Returns:
+            `Tuple(str)`: Paths to the files saved.
+        """
+        if not os.path.isdir(save_directory):
+            logger.error(f"Vocabulary path ({save_directory}) should be a directory")
+            return
+        out_vocab_file = os.path.join(
+            save_directory, (filename_prefix + "-" if filename_prefix else "") + VOCAB_FILES_NAMES["vocab_file"]
+        )
+
+        if os.path.abspath(self.vocab_file) != os.path.abspath(out_vocab_file) and os.path.isfile(self.vocab_file):
+            copyfile(self.vocab_file, out_vocab_file)
+        elif not os.path.isfile(self.vocab_file):
+            with open(out_vocab_file, "wb") as fi:
+                content_spiece_model = self.sp_model.serialized_model_proto()
+                fi.write(content_spiece_model)
+
+        return (out_vocab_file,)
+
+    def build_inputs_with_special_tokens(self, token_ids_0, token_ids_1=None):
+        if self.add_bos_token:
+            bos_token_ids = [self.bos_token_id]
+        else:
+            bos_token_ids = []
+
+        output = bos_token_ids + token_ids_0
+
+        if token_ids_1 is not None:
+            output = output + token_ids_1
+
+        if self.add_eos_token:
+            output = output + [self.eos_token_id]
+
+        return output
+
+    def get_special_tokens_mask(
+        self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None, already_has_special_tokens: bool = False
+    ) -> List[int]:
+        """
+        Retrieve sequence ids from a token list that has no special tokens added. This method is called when adding
+        special tokens using the tokenizer `prepare_for_model` method.
+
+        Args:
+            token_ids_0 (`List[int]`):
+                List of IDs.
+            token_ids_1 (`List[int]`, *optional*):
+                Optional second list of IDs for sequence pairs.
+            already_has_special_tokens (`bool`, *optional*, defaults to `False`):
+                Whether or not the token list is already formatted with special tokens for the model.
+
+        Returns:
+            `List[int]`: A list of integers in the range [0, 1]: 1 for a special token, 0 for a sequence token.
+        """
+        if already_has_special_tokens:
+            return super().get_special_tokens_mask(
+                token_ids_0=token_ids_0, token_ids_1=token_ids_1, already_has_special_tokens=True
+            )
+
+        if token_ids_1 is None:
+            return [1] + ([0] * len(token_ids_0)) + [1]
+        return [1] + ([0] * len(token_ids_0)) + [1, 1] + ([0] * len(token_ids_1)) + [1]
+
+    def create_token_type_ids_from_sequences(
+        self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None
+    ) -> List[int]:
+        """
+        Create a mask from the two sequences passed to be used in a sequence-pair classification task. T5 does not make
+        use of token type ids, therefore a list of zeros is returned.
+
+        Args:
+            token_ids_0 (`List[int]`):
+                List of IDs.
+            token_ids_1 (`List[int]`, *optional*):
+                Optional second list of IDs for sequence pairs.
+
+        Returns:
+            `List[int]`: List of zeros.
+        """
+        eos = [self.eos_token_id]
+
+        if token_ids_1 is None:
+            return len(token_ids_0 + eos) * [0]
+        return len(token_ids_0 + eos + token_ids_1 + eos) * [0]