Mixtral official (#942)

* multipack support for official mixtral implementation

* fix patch to load multipack for mixtral

* chore: lint

examples/mistral/mixtral.yml

@@ -1,5 +1,5 @@
-base_model: DiscoResearch/mixtral-7b-8expert
-model_type: MixtralForCausalLM
+base_model: mistralai/Mixtral-8x7B-v0.1
+model_type: AutoModelForCausalLM
 tokenizer_type: LlamaTokenizer
 trust_remote_code: true
 

requirements.txt

@@ -2,7 +2,7 @@
 auto-gptq==0.5.1
 packaging
 peft==0.6.0
-transformers @ git+https://github.com/huggingface/transformers.git@df5c5c62ae253055336f5bb0828ca8e3e15ab6bd
+transformers @ git+https://github.com/huggingface/transformers.git@e5079b0b2abcef11ecbdae60ba4a6636c57b725d
 tokenizers==0.15.0
 bitsandbytes>=0.41.1
 accelerate==0.24.1

src/axolotl/models/mixtral/__init__.py

@@ -1,6 +0,0 @@
-"""
-Custom modeling code for mixtral
-"""
-
-from .configuration_moe_mistral import MixtralConfig  # noqa
-from .modeling_moe_mistral import MixtralForCausalLM  # noqa

src/axolotl/models/mixtral/configuration_moe_mistral.py

@@ -1,154 +0,0 @@
-# coding=utf-8
-# Copyright 2023 Mistral 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.
-""" Mistral model configuration"""
-
-from transformers.configuration_utils import PretrainedConfig
-from transformers.utils import logging
-
-logger = logging.get_logger(__name__)
-
-MISTRAL_PRETRAINED_CONFIG_ARCHIVE_MAP = {
-    "mistralai/Mistral-7B-v0.1": "https://huggingface.co/mistralai/Mistral-7B-v0.1/resolve/main/config.json",
-    "mistralai/Mistral-7B-Instruct-v0.1": "https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.1/resolve/main/config.json",
-}
-
-
-class MixtralConfig(PretrainedConfig):
-    r"""
-    This is the configuration class to store the configuration of a [`MistralModel`]. It is used to instantiate an
-    Mistral 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 Mistral-7B-v0.1 or Mistral-7B-Instruct-v0.1.
-
-    [mistralai/Mistral-7B-v0.1](https://huggingface.co/mistralai/Mistral-7B-v0.1)
-    [mistralai/Mistral-7B-Instruct-v0.1](https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.1)
-
-    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 Mistral model. Defines the number of different tokens that can be represented by the
-            `inputs_ids` passed when calling [`MistralModel`]
-        hidden_size (`int`, *optional*, defaults to 4096):
-            Dimension of the hidden representations.
-        intermediate_size (`int`, *optional*, defaults to 14336):
-            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.
-        num_key_value_heads (`int`, *optional*, defaults to 8):
-            This is the number of key_value heads that should be used to implement Grouped Query Attention. If
-            `num_key_value_heads=num_attention_heads`, the model will use Multi Head Attention (MHA), if
-            `num_key_value_heads=1 the model will use Multi Query Attention (MQA) otherwise GQA is used. When
-            converting a multi-head checkpoint to a GQA checkpoint, each group key and value head should be constructed
-            by meanpooling all the original heads within that group. For more details checkout [this
-            paper](https://arxiv.org/pdf/2305.13245.pdf). If it is not specified, will default to `8`.
-        hidden_act (`str` or `function`, *optional*, defaults to `"silu"`):
-            The non-linear activation function (function or string) in the decoder.
-        max_position_embeddings (`int`, *optional*, defaults to `4096*32`):
-            The maximum sequence length that this model might ever be used with. Mistral's sliding window attention
-            allows sequence of up to 4096*32 tokens.
-        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-06):
-            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`.
-        pad_token_id (`int`, *optional*):
-            The id of the padding token.
-        bos_token_id (`int`, *optional*, defaults to 1):
-            The id of the "beginning-of-sequence" token.
-        eos_token_id (`int`, *optional*, defaults to 2):
-            The id of the "end-of-sequence" token.
-        tie_word_embeddings (`bool`, *optional*, defaults to `False`):
-            Whether the model's input and output word embeddings should be tied.
-        rope_theta (`float`, *optional*, defaults to 10000.0):
-            The base period of the RoPE embeddings.
-        sliding_window (`int`, *optional*, defaults to 4096):
-            Sliding window attention window size. If not specified, will default to `4096`.
-        attention_dropout (`float`, *optional*, defaults to 0.0):
-            The dropout ratio for the attention probabilities.
-
-    ```python
-    >>> from transformers import MistralModel, MistralConfig
-
-    >>> # Initializing a Mistral 7B style configuration
-    >>> configuration = MixtralConfig()
-
-    >>> # Initializing a model from the Mistral 7B style configuration
-    >>> model = MixtralModel(configuration)
-
-    >>> # Accessing the model configuration
-    >>> configuration = model.config
-    ```"""
-
-    model_type = "mistral"
-    keys_to_ignore_at_inference = ["past_key_values"]
-
-    def __init__(
-        self,
-        vocab_size=32000,
-        hidden_size=4096,
-        intermediate_size=14336,
-        num_hidden_layers=32,
-        num_attention_heads=32,
-        num_key_value_heads=8,
-        hidden_act="silu",
-        max_position_embeddings=4096 * 32,
-        initializer_range=0.02,
-        rms_norm_eps=1e-6,
-        use_cache=True,
-        pad_token_id=None,
-        bos_token_id=1,
-        eos_token_id=2,
-        tie_word_embeddings=False,
-        rope_theta=10000.0,
-        attention_dropout=0.0,
-        num_experts_per_token=2,
-        num_experts=8,
-        **kwargs,
-    ):
-        self.vocab_size = vocab_size
-        self.max_position_embeddings = max_position_embeddings
-        self.hidden_size = hidden_size
-        self.intermediate_size = intermediate_size
-        self.num_hidden_layers = num_hidden_layers
-        self.num_attention_heads = num_attention_heads
-
-        # for backward compatibility
-        if num_key_value_heads is None:
-            num_key_value_heads = num_attention_heads
-
-        self.num_key_value_heads = num_key_value_heads
-        self.hidden_act = hidden_act
-        self.initializer_range = initializer_range
-        self.rms_norm_eps = rms_norm_eps
-        self.use_cache = use_cache
-        self.rope_theta = rope_theta
-        self.attention_dropout = attention_dropout
-        self.num_experts = num_experts
-        self.num_experts_per_token = num_experts_per_token
-
-        # pylint: disable=duplicate-code
-        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,
-        )

src/axolotl/models/mixtral/modeling_moe_mistral.py

@@ -1,1505 +0,0 @@
-# pylint: skip-file
-# coding=utf-8
-# Copyright 2023 Mistral AI and the HuggingFace Inc. team. All rights reserved.
-#
-# This code is based on EleutherAI's GPT-NeoX library and the GPT-NeoX
-# and OPT implementations in this library. It has been modified from its
-# original forms to accommodate minor architectural differences compared
-# to GPT-NeoX and OPT used by the Meta AI team that trained the model.
-#
-# 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 Mistral model."""
-import inspect
-import math
-import warnings
-from typing import List, Optional, Tuple, Union
-
-import torch
-import torch.nn.functional as F
-import torch.utils.checkpoint
-from einops import rearrange
-from torch import nn
-from torch.nn import BCEWithLogitsLoss, CrossEntropyLoss, MSELoss
-from transformers.cache_utils import Cache, DynamicCache
-from transformers.modeling_attn_mask_utils import _prepare_4d_causal_attention_mask
-from transformers.modeling_outputs import (
-    BaseModelOutputWithPast,
-    CausalLMOutputWithPast,
-    SequenceClassifierOutputWithPast,
-)
-from transformers.modeling_utils import PreTrainedModel
-from transformers.utils import (
-    add_start_docstrings,
-    add_start_docstrings_to_model_forward,
-    is_flash_attn_2_available,
-    is_flash_attn_greater_or_equal_2_10,
-    logging,
-    replace_return_docstrings,
-)
-
-from ...monkeypatch.utils import get_cu_seqlens_from_pos_ids
-from .configuration_moe_mistral import MixtralConfig
-
-if is_flash_attn_2_available():
-    from flash_attn import (
-        flash_attn_func,
-        flash_attn_varlen_func,
-        flash_attn_varlen_qkvpacked_func,
-    )
-    from flash_attn.bert_padding import index_first_axis, pad_input, unpad_input  # noqa
-
-    _flash_supports_window_size = "window_size" in list(
-        inspect.signature(flash_attn_func).parameters
-    )
-
-
-logger = logging.get_logger(__name__)
-
-_CONFIG_FOR_DOC = "MixtralConfig"
-
-
-# Copied from transformers.models.llama.modeling_llama._get_unpad_data
-def _get_unpad_data(attention_mask):
-    seqlens_in_batch = attention_mask.sum(dim=-1, dtype=torch.int32)
-    indices = torch.nonzero(attention_mask.flatten(), as_tuple=False).flatten()
-    max_seqlen_in_batch = seqlens_in_batch.max().item()
-    cu_seqlens = F.pad(
-        torch.cumsum(seqlens_in_batch, dim=0, dtype=torch.torch.int32), (1, 0)
-    )
-    return (
-        indices,
-        cu_seqlens,
-        max_seqlen_in_batch,
-    )
-
-
-# Copied from transformers.models.llama.modeling_llama.LlamaRMSNorm with Llama->Mistral
-class MistralRMSNorm(nn.Module):
-    def __init__(self, hidden_size, eps=1e-6):
-        """
-        MistralRMSNorm is equivalent to T5LayerNorm
-        """
-        super().__init__()
-        self.weight = nn.Parameter(torch.ones(hidden_size))
-        self.variance_epsilon = eps
-
-    def forward(self, hidden_states):
-        input_dtype = hidden_states.dtype
-        hidden_states = hidden_states.to(torch.float32)
-        variance = hidden_states.pow(2).mean(-1, keepdim=True)
-        hidden_states = hidden_states * torch.rsqrt(variance + self.variance_epsilon)
-        return self.weight * hidden_states.to(input_dtype)
-
-
-# Copied from transformers.models.llama.modeling_llama.LlamaRotaryEmbedding with Llama->Mistral
-class MistralRotaryEmbedding(nn.Module):
-    def __init__(self, dim, max_position_embeddings=2048, base=10000, device=None):
-        super().__init__()
-
-        self.dim = dim
-        self.max_position_embeddings = max_position_embeddings
-        self.base = base
-        inv_freq = 1.0 / (
-            self.base ** (torch.arange(0, self.dim, 2).float().to(device) / self.dim)
-        )
-        self.register_buffer("inv_freq", inv_freq, persistent=False)
-
-        # Build here to make `torch.jit.trace` work.
-        self._set_cos_sin_cache(
-            seq_len=max_position_embeddings,
-            device=self.inv_freq.device,
-            dtype=torch.get_default_dtype(),
-        )
-
-    def _set_cos_sin_cache(self, seq_len, device, dtype):
-        self.max_seq_len_cached = seq_len
-        t = torch.arange(
-            self.max_seq_len_cached, device=device, dtype=self.inv_freq.dtype
-        )
-
-        freqs = torch.outer(t, self.inv_freq)
-        # Different from paper, but it uses a different permutation in order to obtain the same calculation
-        emb = torch.cat((freqs, freqs), dim=-1)
-        self.register_buffer("cos_cached", emb.cos().to(dtype), persistent=False)
-        self.register_buffer("sin_cached", emb.sin().to(dtype), persistent=False)
-
-    def forward(self, x, seq_len=None):
-        # x: [bs, num_attention_heads, seq_len, head_size]
-        if seq_len > self.max_seq_len_cached:
-            self._set_cos_sin_cache(seq_len=seq_len, device=x.device, dtype=x.dtype)
-
-        return (
-            self.cos_cached[:seq_len].to(dtype=x.dtype),
-            self.sin_cached[:seq_len].to(dtype=x.dtype),
-        )
-
-
-# Copied from transformers.models.llama.modeling_llama.rotate_half
-def rotate_half(x):
-    """Rotates half the hidden dims of the input."""
-    x1 = x[..., : x.shape[-1] // 2]
-    x2 = x[..., x.shape[-1] // 2 :]
-    return torch.cat((-x2, x1), dim=-1)
-
-
-# Copied from transformers.models.llama.modeling_llama.apply_rotary_pos_emb
-def apply_rotary_pos_emb(q, k, cos, sin, position_ids, unsqueeze_dim=1):
-    """Applies Rotary Position Embedding to the query and key tensors.
-
-    Args:
-        q (`torch.Tensor`): The query tensor.
-        k (`torch.Tensor`): The key tensor.
-        cos (`torch.Tensor`): The cosine part of the rotary embedding.
-        sin (`torch.Tensor`): The sine part of the rotary embedding.
-        position_ids (`torch.Tensor`):
-            The position indices of the tokens corresponding to the query and key tensors. For example, this can be
-            used to pass offsetted position ids when working with a KV-cache.
-        unsqueeze_dim (`int`, *optional*, defaults to 1):
-            The 'unsqueeze_dim' argument specifies the dimension along which to unsqueeze cos[position_ids] and
-            sin[position_ids] so that they can be properly broadcasted to the dimensions of q and k. For example, note
-            that cos[position_ids] and sin[position_ids] have the shape [batch_size, seq_len, head_dim]. Then, if q and
-            k have the shape [batch_size, heads, seq_len, head_dim], then setting unsqueeze_dim=1 makes
-            cos[position_ids] and sin[position_ids] broadcastable to the shapes of q and k. Similarly, if q and k have
-            the shape [batch_size, seq_len, heads, head_dim], then set unsqueeze_dim=2.
-    Returns:
-        `tuple(torch.Tensor)` comprising of the query and key tensors rotated using the Rotary Position Embedding.
-    """
-    cos = cos[position_ids].unsqueeze(unsqueeze_dim)
-    sin = sin[position_ids].unsqueeze(unsqueeze_dim)
-    q_embed = (q * cos) + (rotate_half(q) * sin)
-    k_embed = (k * cos) + (rotate_half(k) * sin)
-    return q_embed, k_embed
-
-
-class FeedForward(nn.Module):
-    def __init__(self, config):
-        """
-        Initialize the FeedForward module.
-
-        Args:
-            dim (int): Input dimension.
-            hidden_dim (int): Hidden dimension of the feedforward layer.
-            multiple_of (int): Value to ensure hidden dimension is a multiple of this value.
-            ffn_dim_multiplier (float, optional): Custom multiplier for hidden dimension. Defaults to None.
-
-        Attributes:
-            w1 (ColumnParallelLinear): Linear transformation for the first layer.
-            w2 (RowParallelLinear): Linear transformation for the second layer.
-            w3 (ColumnParallelLinear): Linear transformation for the third layer.
-
-        """
-        super().__init__()
-
-        self.w1 = nn.Linear(config.hidden_size, config.intermediate_size, bias=False)
-        self.w2 = nn.Linear(config.intermediate_size, config.hidden_size, bias=False)
-        self.w3 = nn.Linear(config.hidden_size, config.intermediate_size, bias=False)
-
-    def forward(self, x):
-        return self.w2(F.silu(self.w1(x)) * self.w3(x))
-
-
-class MoE(nn.Module):
-    def __init__(
-        self,
-        config,
-    ):
-        super().__init__()
-        self.config = config
-        self.gate = nn.Linear(config.hidden_size, config.num_experts, bias=False)
-        self.experts = nn.ModuleList(
-            [FeedForward(config) for i in range(config.num_experts)]
-        )
-
-    def forward(self, x):
-        orig_shape = x.shape
-        x = x.view(-1, x.shape[-1])
-
-        scores = self.gate(x).softmax(dim=-1)
-        expert_weights, expert_indices = torch.topk(
-            scores, self.config.num_experts_per_token, dim=-1
-        )
-        flat_expert_indices = expert_indices.view(-1)
-
-        x = x.repeat_interleave(self.config.num_experts_per_token, dim=0)
-        y = torch.empty_like(x)
-        for i, expert in enumerate(self.experts):
-            y[flat_expert_indices == i] = expert(x[flat_expert_indices == i])
-        y = (y.view(*expert_weights.shape, -1) * expert_weights.unsqueeze(-1)).sum(
-            dim=1
-        )
-        return y.view(*orig_shape)
-
-
-# Copied from transformers.models.llama.modeling_llama.repeat_kv
-def repeat_kv(hidden_states: torch.Tensor, n_rep: int) -> torch.Tensor:
-    """
-    This is the equivalent of torch.repeat_interleave(x, dim=1, repeats=n_rep). The hidden states go from (batch,
-    num_key_value_heads, seqlen, head_dim) to (batch, num_attention_heads, seqlen, head_dim)
-    """
-    batch, num_key_value_heads, slen, head_dim = hidden_states.shape
-    if n_rep == 1:
-        return hidden_states
-    hidden_states = hidden_states[:, :, None, :, :].expand(
-        batch, num_key_value_heads, n_rep, slen, head_dim
-    )
-    return hidden_states.reshape(batch, num_key_value_heads * n_rep, slen, head_dim)
-
-
-class MistralAttention(nn.Module):
-    """
-    Multi-headed attention from 'Attention Is All You Need' paper. Modified to use sliding window attention: Longformer
-    and "Generating Long Sequences with Sparse Transformers".
-    """
-
-    def __init__(self, config: MixtralConfig, layer_idx: Optional[int] = None):
-        super().__init__()
-        self.config = config
-        self.layer_idx = layer_idx
-        if layer_idx is None:
-            logger.warning_once(
-                f"Instantiating {self.__class__.__name__} without passing `layer_idx` is not recommended and will "
-                "to errors during the forward call, if caching is used. Please make sure to provide a `layer_idx` "
-                "when creating this class."
-            )
-
-        self.hidden_size = config.hidden_size
-        self.num_heads = config.num_attention_heads
-        self.head_dim = self.hidden_size // self.num_heads
-        self.num_key_value_heads = config.num_key_value_heads
-        self.num_key_value_groups = self.num_heads // self.num_key_value_heads
-        self.max_position_embeddings = config.max_position_embeddings
-        self.rope_theta = config.rope_theta
-        self.is_causal = True
-        self.attention_dropout = config.attention_dropout
-
-        if (self.head_dim * self.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`: {self.num_heads})."
-            )
-        self.q_proj = nn.Linear(
-            self.hidden_size, self.num_heads * self.head_dim, bias=False
-        )
-        self.k_proj = nn.Linear(
-            self.hidden_size, self.num_key_value_heads * self.head_dim, bias=False
-        )
-        self.v_proj = nn.Linear(
-            self.hidden_size, self.num_key_value_heads * self.head_dim, bias=False
-        )
-        self.o_proj = nn.Linear(
-            self.num_heads * self.head_dim, self.hidden_size, bias=False
-        )
-
-        self.rotary_emb = MistralRotaryEmbedding(
-            self.head_dim,
-            max_position_embeddings=self.max_position_embeddings,
-            base=self.rope_theta,
-        )
-
-    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,
-        attention_mask: Optional[torch.Tensor] = None,
-        position_ids: Optional[torch.LongTensor] = None,
-        past_key_value: Optional[Cache] = None,
-        output_attentions: bool = False,
-        use_cache: bool = False,
-        **kwargs,
-    ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
-        if "padding_mask" in kwargs:
-            warnings.warn(
-                "Passing `padding_mask` is deprecated and will be removed in v4.37. Please make sure use `attention_mask` instead.`"
-            )
-        bsz, q_len, _ = hidden_states.size()
-
-        query_states = self.q_proj(hidden_states)
-        key_states = self.k_proj(hidden_states)
-        value_states = self.v_proj(hidden_states)
-
-        query_states = query_states.view(
-            bsz, q_len, self.num_heads, self.head_dim
-        ).transpose(1, 2)
-        key_states = key_states.view(
-            bsz, q_len, self.num_key_value_heads, self.head_dim
-        ).transpose(1, 2)
-        value_states = value_states.view(
-            bsz, q_len, self.num_key_value_heads, self.head_dim
-        ).transpose(1, 2)
-
-        kv_seq_len = key_states.shape[-2]
-        if past_key_value is not None:
-            if self.layer_idx is None:
-                raise ValueError(
-                    f"The cache structure has changed since version v4.36. If you are using {self.__class__.__name__} "
-                    "for auto-regressive decoding with k/v caching, please make sure to initialize the attention class "
-                    "with a layer index."
-                )
-            kv_seq_len += past_key_value.get_seq_length(self.layer_idx)
-        cos, sin = self.rotary_emb(value_states, seq_len=kv_seq_len)
-        query_states, key_states = apply_rotary_pos_emb(
-            query_states, key_states, cos, sin, position_ids
-        )
-
-        if past_key_value is not None:
-            cache_kwargs = {"sin": sin, "cos": cos}  # Specific to RoPE models
-            key_states, value_states = past_key_value.update(
-                key_states, value_states, self.layer_idx, cache_kwargs
-            )
-
-        # repeat k/v heads if n_kv_heads < n_heads
-        key_states = repeat_kv(key_states, self.num_key_value_groups)
-        value_states = repeat_kv(value_states, self.num_key_value_groups)
-
-        attn_weights = torch.matmul(
-            query_states, key_states.transpose(2, 3)
-        ) / math.sqrt(self.head_dim)
-
-        if attn_weights.size() != (bsz, self.num_heads, q_len, kv_seq_len):
-            raise ValueError(
-                f"Attention weights should be of size {(bsz, self.num_heads, q_len, kv_seq_len)}, but is"
-                f" {attn_weights.size()}"
-            )
-
-        if attention_mask is not None:
-            if attention_mask.size() != (bsz, 1, q_len, kv_seq_len):
-                raise ValueError(
-                    f"Attention mask should be of size {(bsz, 1, q_len, kv_seq_len)}, but is {attention_mask.size()}"
-                )
-
-            attn_weights = attn_weights + attention_mask
-
-        # upcast attention to fp32
-        attn_weights = nn.functional.softmax(
-            attn_weights, dim=-1, dtype=torch.float32
-        ).to(query_states.dtype)
-        attn_weights = nn.functional.dropout(
-            attn_weights, p=self.attention_dropout, training=self.training
-        )
-        attn_output = torch.matmul(attn_weights, value_states)
-
-        if attn_output.size() != (bsz, self.num_heads, q_len, self.head_dim):
-            raise ValueError(
-                f"`attn_output` should be of size {(bsz, self.num_heads, q_len, self.head_dim)}, but is"
-                f" {attn_output.size()}"
-            )
-
-        attn_output = attn_output.transpose(1, 2).contiguous()
-        attn_output = attn_output.reshape(bsz, q_len, self.hidden_size)
-
-        attn_output = self.o_proj(attn_output)
-
-        if not output_attentions:
-            attn_weights = None
-
-        return attn_output, attn_weights, past_key_value
-
-
-class MistralFlashAttention2(MistralAttention):
-    """
-    Mistral flash attention module. This module inherits from `MistralAttention` as the weights of the module stays
-    untouched. The only required change would be on the forward pass where it needs to correctly call the public API of
-    flash attention and deal with padding tokens in case the input contains any of them.
-    """
-
-    # Copied from transformers.models.llama.modeling_llama.LlamaFlashAttention2.__init__
-    def __init__(self, *args, **kwargs):
-        super().__init__(*args, **kwargs)
-
-        # TODO: Should be removed once Flash Attention for RoCm is bumped to 2.1.
-        # flash_attn<2.1 generates top-left aligned causal mask, while what is needed here is bottom-right alignement, that was made default for flash_attn>=2.1. This attribute is used to handle this difference. Reference: https://github.com/Dao-AILab/flash-attention/releases/tag/v2.1.0.
-        # Beware that with flash_attn<2.1, using q_seqlen != k_seqlen (except for the case q_seqlen == 1) produces a wrong mask (top-left).
-        self._flash_attn_uses_top_left_mask = not is_flash_attn_greater_or_equal_2_10()
-
-    def forward(
-        self,
-        hidden_states: torch.Tensor,
-        attention_mask: Optional[torch.Tensor] = None,
-        position_ids: Optional[torch.LongTensor] = None,
-        past_key_value: Optional[Cache] = None,
-        output_attentions: bool = False,
-        use_cache: bool = False,
-        cu_seqlens: Optional[torch.Tensor] = None,
-        max_seqlen: Optional[torch.Tensor] = None,
-        **kwargs,
-    ):
-        if "padding_mask" in kwargs:
-            warnings.warn(
-                "Passing `padding_mask` is deprecated and will be removed in v4.37. Please make sure use `attention_mask` instead.`"
-            )
-
-            # overwrite attention_mask with padding_mask
-            attention_mask = kwargs.pop("padding_mask")
-        bsz, q_len, _ = hidden_states.size()
-
-        query_states = self.q_proj(hidden_states)
-        key_states = self.k_proj(hidden_states)
-        value_states = self.v_proj(hidden_states)
-
-        query_states = query_states.view(
-            bsz, q_len, self.num_heads, self.head_dim
-        ).transpose(1, 2)
-        key_states = key_states.view(
-            bsz, q_len, self.num_key_value_heads, self.head_dim
-        ).transpose(1, 2)
-        value_states = value_states.view(
-            bsz, q_len, self.num_key_value_heads, self.head_dim
-        ).transpose(1, 2)
-
-        kv_seq_len = key_states.shape[-2]
-        if past_key_value is not None:
-            kv_seq_len += past_key_value.get_seq_length(self.layer_idx)
-
-        # Because the input can be padded, the absolute sequence length depends on the max position id.
-        rotary_seq_len = max(kv_seq_len, position_ids[:, -1].max().item()) + 1
-        cos, sin = self.rotary_emb(value_states, seq_len=rotary_seq_len)
-
-        query_states, key_states = apply_rotary_pos_emb(
-            query_states, key_states, cos, sin, position_ids
-        )
-
-        use_sliding_windows = (
-            _flash_supports_window_size
-            and getattr(self.config, "sliding_window", None) is not None
-            and kv_seq_len > self.config.sliding_window
-        )
-
-        if not _flash_supports_window_size:
-            logger.warning_once(
-                "The current flash attention version does not support sliding window attention, for a more memory efficient implementation"
-                " make sure to upgrade flash-attn library."
-            )
-
-        if past_key_value is not None:
-            # Activate slicing cache only if the config has a value `sliding_windows` attribute
-            if (
-                getattr(self.config, "sliding_window", None) is not None
-                and kv_seq_len > self.config.sliding_window
-            ):
-                slicing_tokens = 1 - self.config.sliding_window
-
-                past_key = past_key_value[0]
-                past_value = past_key_value[1]
-
-                past_key = past_key[:, :, slicing_tokens:, :].contiguous()
-                past_value = past_value[:, :, slicing_tokens:, :].contiguous()
-
-                if past_key.shape[-2] != self.config.sliding_window - 1:
-                    raise ValueError(
-                        f"past key must have a shape of (`batch_size, num_heads, self.config.sliding_window-1, head_dim`), got"
-                        f" {past_key.shape}"
-                    )
-
-                past_key_value = (past_key, past_value)
-
-                if attention_mask is not None:
-                    attention_mask = attention_mask[:, slicing_tokens:]
-                    attention_mask = torch.cat(
-                        [attention_mask, torch.ones_like(attention_mask[:, -1:])],
-                        dim=-1,
-                    )
-
-            cache_kwargs = {"sin": sin, "cos": cos}  # Specific to RoPE models
-            key_states, value_states = past_key_value.update(
-                key_states, value_states, self.layer_idx, cache_kwargs
-            )
-
-        # repeat k/v heads if n_kv_heads < n_heads
-        key_states = repeat_kv(key_states, self.num_key_value_groups)
-        value_states = repeat_kv(value_states, self.num_key_value_groups)
-        dropout_rate = 0.0 if not self.training else self.attention_dropout
-
-        if cu_seqlens is not None and max_seqlen is not None and cu_seqlens.dim() == 1:
-            # special handling using sample packing
-            qkv = torch.stack(
-                [query_states, key_states, value_states], dim=2
-            )  # [bsz, nh, 3, q_len, hd]
-            qkv = qkv.transpose(1, 3)  # [bsz, q_len, 3, nh, hd]
-            qkv = rearrange(qkv, "b s ... -> (b s) ...")
-
-            attn_output = flash_attn_varlen_qkvpacked_func(
-                qkv,
-                cu_seqlens,
-                max_seqlen,
-                dropout_p=dropout_rate,
-                softmax_scale=None,
-                causal=True,
-            )
-            attn_output = rearrange(attn_output, "(b s) ... -> b s ...", b=bsz)
-        else:
-            # In PEFT, usually we cast the layer norms in float32 for training stability reasons
-            # therefore the input hidden states gets silently casted in float32. Hence, we need
-            # cast them back in float16 just to be sure everything works as expected.
-            input_dtype = query_states.dtype
-            if input_dtype == torch.float32:
-                # Handle the case where the model is quantized
-                if hasattr(self.config, "_pre_quantization_dtype"):
-                    target_dtype = self.config._pre_quantization_dtype
-                else:
-                    target_dtype = self.q_proj.weight.dtype
-
-                logger.warning_once(
-                    f"The input hidden states seems to be silently casted in float32, this might be related to"
-                    f" the fact you have upcasted embedding or layer norm layers in float32. We will cast back the input in"
-                    f" {target_dtype}."
-                )
-
-                query_states = query_states.to(target_dtype)
-                key_states = key_states.to(target_dtype)
-                value_states = value_states.to(target_dtype)
-
-            # Reashape to the expected shape for Flash Attention
-            query_states = query_states.transpose(1, 2)
-            key_states = key_states.transpose(1, 2)
-            value_states = value_states.transpose(1, 2)
-
-            attn_output = self._flash_attention_forward(
-                query_states,
-                key_states,
-                value_states,
-                attention_mask,
-                q_len,
-                dropout=dropout_rate,
-                use_sliding_windows=use_sliding_windows,
-            )
-
-        attn_output = attn_output.reshape(bsz, q_len, self.hidden_size).contiguous()
-        attn_output = self.o_proj(attn_output)
-
-        if not output_attentions:
-            attn_weights = None
-
-        return attn_output, attn_weights, past_key_value
-
-    def _flash_attention_forward(
-        self,
-        query_states,
-        key_states,
-        value_states,
-        attention_mask,
-        query_length,
-        dropout=0.0,
-        softmax_scale=None,
-        use_sliding_windows=False,
-    ):
-        """
-        Calls the forward method of Flash Attention - if the input hidden states contain at least one padding token
-        first unpad the input, then computes the attention scores and pad the final attention scores.
-
-        Args:
-            query_states (`torch.Tensor`):
-                Input query states to be passed to Flash Attention API
-            key_states (`torch.Tensor`):
-                Input key states to be passed to Flash Attention API
-            value_states (`torch.Tensor`):
-                Input value states to be passed to Flash Attention API
-            attention_mask (`torch.Tensor`):
-                The padding mask - corresponds to a tensor of size `(batch_size, seq_len)` where 0 stands for the
-                position of padding tokens and 1 for the position of non-padding tokens.
-            dropout (`int`, *optional*):
-                Attention dropout
-            softmax_scale (`float`, *optional*):
-                The scaling of QK^T before applying softmax. Default to 1 / sqrt(head_dim)
-            use_sliding_windows (`bool`, *optional*):
-                Whether to activate sliding window attention.
-        """
-        if not self._flash_attn_uses_top_left_mask:
-            causal = self.is_causal
-        else:
-            # TODO: Remove the `query_length != 1` check once Flash Attention for RoCm is bumped to 2.1. For details, please see the comment in LlamaFlashAttention2 __init__.
-            causal = self.is_causal and query_length != 1
-
-        # Contains at least one padding token in the sequence
-        if attention_mask is not None:
-            batch_size = query_states.shape[0]
-            (
-                query_states,
-                key_states,
-                value_states,
-                indices_q,
-                cu_seq_lens,
-                max_seq_lens,
-            ) = self._upad_input(
-                query_states, key_states, value_states, attention_mask, query_length
-            )
-
-            cu_seqlens_q, cu_seqlens_k = cu_seq_lens
-            max_seqlen_in_batch_q, max_seqlen_in_batch_k = max_seq_lens
-
-            if not use_sliding_windows:
-                attn_output_unpad = flash_attn_varlen_func(
-                    query_states,
-                    key_states,
-                    value_states,
-                    cu_seqlens_q=cu_seqlens_q,
-                    cu_seqlens_k=cu_seqlens_k,
-                    max_seqlen_q=max_seqlen_in_batch_q,
-                    max_seqlen_k=max_seqlen_in_batch_k,
-                    dropout_p=dropout,
-                    softmax_scale=softmax_scale,
-                    causal=causal,
-                )
-            else:
-                attn_output_unpad = flash_attn_varlen_func(
-                    query_states,
-                    key_states,
-                    value_states,
-                    cu_seqlens_q=cu_seqlens_q,
-                    cu_seqlens_k=cu_seqlens_k,
-                    max_seqlen_q=max_seqlen_in_batch_q,
-                    max_seqlen_k=max_seqlen_in_batch_k,
-                    dropout_p=dropout,
-                    softmax_scale=softmax_scale,
-                    causal=causal,
-                    window_size=(
-                        self.config.sliding_window,
-                        self.config.sliding_window,
-                    ),
-                )
-
-            attn_output = pad_input(
-                attn_output_unpad, indices_q, batch_size, query_length
-            )
-        else:
-            if not use_sliding_windows:
-                attn_output = flash_attn_func(
-                    query_states,
-                    key_states,
-                    value_states,
-                    dropout,
-                    softmax_scale=softmax_scale,
-                    causal=causal,
-                )
-            else:
-                attn_output = flash_attn_func(
-                    query_states,
-                    key_states,
-                    value_states,
-                    dropout,
-                    softmax_scale=softmax_scale,
-                    causal=causal,
-                    window_size=(
-                        self.config.sliding_window,
-                        self.config.sliding_window,
-                    ),
-                )
-
-        return attn_output
-
-    def _upad_input(
-        self, query_layer, key_layer, value_layer, attention_mask, query_length
-    ):
-        batch_size, kv_seq_len, num_heads, head_dim = key_layer.shape
-
-        # On the first iteration we need to properly re-create the padding mask
-        # by slicing it on the proper place
-        if kv_seq_len != attention_mask.shape[-1]:
-            attention_mask_num_tokens = attention_mask.shape[-1]
-            attention_mask = attention_mask[:, attention_mask_num_tokens - kv_seq_len :]
-
-        indices_k, cu_seqlens_k, max_seqlen_in_batch_k = _get_unpad_data(attention_mask)
-
-        key_layer = index_first_axis(
-            key_layer.reshape(batch_size * kv_seq_len, num_heads, head_dim), indices_k
-        )
-        value_layer = index_first_axis(
-            value_layer.reshape(batch_size * kv_seq_len, num_heads, head_dim), indices_k
-        )
-
-        if query_length == kv_seq_len:
-            query_layer = index_first_axis(
-                query_layer.reshape(batch_size * kv_seq_len, num_heads, head_dim),
-                indices_k,
-            )
-            cu_seqlens_q = cu_seqlens_k
-            max_seqlen_in_batch_q = max_seqlen_in_batch_k
-            indices_q = indices_k
-        elif query_length == 1:
-            max_seqlen_in_batch_q = 1
-            cu_seqlens_q = torch.arange(
-                batch_size + 1, dtype=torch.int32, device=query_layer.device
-            )  # There is a memcpy here, that is very bad.
-            indices_q = cu_seqlens_q[:-1]
-            query_layer = query_layer.squeeze(1)
-        else:
-            # The -q_len: slice assumes left padding.
-            attention_mask = attention_mask[:, -query_length:]
-            query_layer, indices_q, cu_seqlens_q, max_seqlen_in_batch_q = unpad_input(
-                query_layer, attention_mask
-            )
-
-        return (
-            query_layer,
-            key_layer,
-            value_layer,
-            indices_q,
-            (cu_seqlens_q, cu_seqlens_k),
-            (max_seqlen_in_batch_q, max_seqlen_in_batch_k),
-        )
-
-
-class MixtralDecoderLayer(nn.Module):
-    def __init__(self, config: MixtralConfig, layer_idx: int):
-        super().__init__()
-        self.hidden_size = config.hidden_size
-        self.self_attn = MistralFlashAttention2(config, layer_idx=layer_idx)
-        self.mlp = MoE(config)
-        self.input_layernorm = MistralRMSNorm(
-            config.hidden_size, eps=config.rms_norm_eps
-        )
-        self.post_attention_layernorm = MistralRMSNorm(
-            config.hidden_size, eps=config.rms_norm_eps
-        )
-
-    def forward(
-        self,
-        hidden_states: torch.Tensor,
-        attention_mask: Optional[torch.Tensor] = None,
-        position_ids: Optional[torch.LongTensor] = None,
-        past_key_value: Optional[Tuple[torch.Tensor]] = None,
-        output_attentions: Optional[bool] = False,
-        use_cache: Optional[bool] = False,
-        cu_seqlens: Optional[torch.Tensor] = None,
-        max_seqlen: Optional[torch.Tensor] = None,
-        **kwargs,
-    ) -> Tuple[
-        torch.FloatTensor, Optional[Tuple[torch.FloatTensor, torch.FloatTensor]]
-    ]:
-        if "padding_mask" in kwargs:
-            warnings.warn(
-                "Passing `padding_mask` is deprecated and will be removed in v4.37. Please make sure use `attention_mask` instead.`"
-            )
-        """
-        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, sequence_length)` where padding elements are indicated by 0.
-            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.input_layernorm(hidden_states)
-
-        # Self Attention
-        # pylint: disable=duplicate-code
-        hidden_states, self_attn_weights, present_key_value = self.self_attn(
-            hidden_states=hidden_states,
-            attention_mask=attention_mask,
-            position_ids=position_ids,
-            past_key_value=past_key_value,
-            output_attentions=output_attentions,
-            use_cache=use_cache,
-            cu_seqlens=cu_seqlens,
-            max_seqlen=max_seqlen,
-        )
-        hidden_states = residual + hidden_states
-
-        # Fully Connected
-        residual = hidden_states
-        hidden_states = self.post_attention_layernorm(hidden_states)
-        hidden_states = self.mlp(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
-
-
-MISTRAL_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 ([`MixtralConfig`]):
-            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 Mistral Model outputting raw hidden-states without any specific head on top.",
-    MISTRAL_START_DOCSTRING,
-)
-class MixtralPreTrainedModel(PreTrainedModel):
-    config_class = MixtralConfig
-    base_model_prefix = "model"
-    supports_gradient_checkpointing = True
-    _no_split_modules = ["MixtralDecoderLayer"]
-    _skip_keys_device_placement = "past_key_values"
-    _supports_flash_attn_2 = True
-    _supports_cache_class = True
-
-    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_()
-
-
-MISTRAL_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.
-
-            - 1 indicates the head is **not masked**,
-            - 0 indicates the head is **masked**.
-        position_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
-            Indices of positions of each input sequence tokens in the position embeddings. Selected in the range `[0,
-            config.n_positions - 1]`.
-
-            [What are position IDs?](../glossary#position-ids)
-        past_key_values (`Cache` or `tuple(tuple(torch.FloatTensor))`, *optional*):
-            Pre-computed hidden-states (key and values in the self-attention blocks and in the cross-attention
-            blocks) that can be used to speed up sequential decoding. This typically consists in the `past_key_values`
-            returned by the model at a previous stage of decoding, when `use_cache=True` or `config.use_cache=True`.
-
-            Two formats are allowed:
-            - a [`~cache_utils.Cache`] instance;
-            - 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)`). This is also known as the legacy
-            cache format.
-
-            The model will output the same cache format that is fed as input. If no `past_key_values` are passed, the
-            legacy cache format will be returned.
-
-            If `past_key_values` are used, the user can optionally input only the last `input_ids` (those that don't
-            have their past key value states given to this model) of shape `(batch_size, 1)` instead of all `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.
-"""
-
-
-@add_start_docstrings(
-    "The bare Mistral Model outputting raw hidden-states without any specific head on top.",
-    MISTRAL_START_DOCSTRING,
-)
-class MistralModel(MixtralPreTrainedModel):
-    """
-    Transformer decoder consisting of *config.num_hidden_layers* layers. Each layer is a [`MixtralDecoderLayer`]
-
-    Args:
-        config: MixtralConfig
-    """
-
-    def __init__(self, config: MixtralConfig):
-        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(
-            [
-                MixtralDecoderLayer(config, layer_idx)
-                for layer_idx in range(config.num_hidden_layers)
-            ]
-        )
-        self._use_flash_attention_2 = config._attn_implementation == "flash_attention_2"
-        self.norm = MistralRMSNorm(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
-
-    @add_start_docstrings_to_model_forward(MISTRAL_INPUTS_DOCSTRING)
-    def forward(
-        self,
-        input_ids: torch.LongTensor = None,
-        attention_mask: Optional[torch.Tensor] = None,
-        position_ids: Optional[torch.LongTensor] = 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
-        )
-
-        # 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:
-            batch_size, seq_length = input_ids.shape
-        elif inputs_embeds is not None:
-            batch_size, seq_length, _ = inputs_embeds.shape
-        else:
-            raise ValueError(
-                "You have to specify either decoder_input_ids or decoder_inputs_embeds"
-            )
-
-        seq_length_with_past = seq_length
-        past_key_values_length = 0
-
-        if use_cache:
-            use_legacy_cache = not isinstance(past_key_values, Cache)
-            if use_legacy_cache:
-                past_key_values = DynamicCache.from_legacy_cache(past_key_values)
-            past_key_values_length = past_key_values.get_seq_length()
-            seq_length_with_past = seq_length_with_past + past_key_values_length
-
-        cu_seqlens = None
-        max_seqlen = None
-        if position_ids is None:
-            device = input_ids.device if input_ids is not None else inputs_embeds.device
-            position_ids = torch.arange(
-                past_key_values_length,
-                seq_length + past_key_values_length,
-                dtype=torch.long,
-                device=device,
-            )
-            position_ids = position_ids.unsqueeze(0).view(-1, seq_length)
-        else:
-            position_ids = position_ids.view(-1, seq_length).long()
-            cu_seqlens, max_seqlen = get_cu_seqlens_from_pos_ids(position_ids)
-            cu_seqlens = cu_seqlens.squeeze()
-
-        if inputs_embeds is None:
-            inputs_embeds = self.embed_tokens(input_ids)
-
-        if (
-            attention_mask is not None
-            and hasattr(self.config, "_flash_attn_2_enabled")
-            and self.config._flash_attn_2_enabled
-            and use_cache
-        ):
-            is_padding_right = attention_mask[:, -1].sum().item() != batch_size
-            if is_padding_right:
-                raise ValueError(
-                    "You are attempting to perform batched generation with padding_side='right'"
-                    " this may lead to unexpected behaviour for Flash Attention version of Mistral. Make sure to "
-                    " call `tokenizer.padding_side  = 'left'` before tokenizing the input. "
-                )
-
-        if getattr(self.config, "_flash_attn_2_enabled", False):
-            # 2d mask is passed through the layers
-            attention_mask = (
-                attention_mask
-                if (attention_mask is not None and 0 in attention_mask)
-                else None
-            )
-        else:
-            # 4d mask is passed through the layers
-            attention_mask = _prepare_4d_causal_attention_mask(
-                attention_mask,
-                (batch_size, seq_length),
-                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 = None
-
-        for decoder_layer in self.layers:
-            if output_hidden_states:
-                all_hidden_states += (hidden_states,)
-
-            if self.gradient_checkpointing and self.training:
-                layer_outputs = self._gradient_checkpointing_func(
-                    decoder_layer.__call__,
-                    hidden_states,
-                    attention_mask,
-                    position_ids,
-                    past_key_values,
-                    output_attentions,
-                    use_cache,
-                    cu_seqlens,
-                    max_seqlen,
-                )
-            else:
-                layer_outputs = decoder_layer(
-                    hidden_states,
-                    attention_mask=attention_mask,
-                    position_ids=position_ids,
-                    past_key_value=past_key_values,
-                    output_attentions=output_attentions,
-                    use_cache=use_cache,
-                    cu_seqlens=cu_seqlens,
-                    max_seqlen=max_seqlen,
-                )
-
-            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 = None
-        if use_cache:
-            next_cache = (
-                next_decoder_cache.to_legacy_cache()
-                if use_legacy_cache
-                else next_decoder_cache
-            )
-
-        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,
-        )
-
-
-class MixtralForCausalLM(MixtralPreTrainedModel):
-    _tied_weights_keys = ["lm_head.weight"]
-
-    def __init__(self, config):
-        super().__init__(config)
-        self.model = MistralModel(config)
-        self.vocab_size = config.vocab_size
-        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.embed_tokens
-
-    def set_input_embeddings(self, value):
-        self.model.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
-
-    def get_decoder(self):
-        return self.model
-
-    def _init_weights(self, module):
-        return
-
-    @add_start_docstrings_to_model_forward(MISTRAL_INPUTS_DOCSTRING)
-    @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,
-        position_ids: Optional[torch.LongTensor] = 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:
-            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]`.
-
-        Returns:
-
-        Example:
-
-        ```python
-        >>> from transformers import AutoTokenizer, MistralForCausalLM
-
-        >>> model = MistralForCausalLM.from_pretrained(PATH_TO_CONVERTED_WEIGHTS)
-        >>> tokenizer = AutoTokenizer.from_pretrained(PATH_TO_CONVERTED_TOKENIZER)
-
-        >>> prompt = "Hey, are you conscious? 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 conscious? Can you talk to me?\nI'm not conscious, 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(
-            input_ids=input_ids,
-            attention_mask=attention_mask,
-            position_ids=position_ids,
-            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,
-        )
-
-        hidden_states = outputs[0]
-        logits = self.lm_head(hidden_states)
-        logits = logits.float()
-
-        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()
-            shift_logits = shift_logits.view(-1, self.config.vocab_size)
-            shift_labels = shift_labels.view(-1)
-            # Enable model parallelism
-            shift_labels = shift_labels.to(shift_logits.device)
-            loss = loss_fct(shift_logits, shift_labels)
-
-        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,
-    ):
-        # Omit tokens covered by past_key_values
-        if past_key_values is not None:
-            if isinstance(past_key_values, Cache):
-                cache_length = past_key_values.get_seq_length()
-                past_length = past_key_values.seen_tokens
-            else:
-                cache_length = past_length = past_key_values[0][0].shape[2]
-
-            # Keep only the unprocessed tokens:
-            # 1 - If the length of the attention_mask exceeds the length of input_ids, then we are in a setting where
-            # some of the inputs are exclusivelly passed as part of the cache (e.g. when passing input_embeds as
-            # input)
-            if (
-                attention_mask is not None
-                and attention_mask.shape[1] > input_ids.shape[1]
-            ):
-                input_ids = input_ids[:, -(attention_mask.shape[1] - past_length) :]
-            # 2 - If the past_length is smaller than input_ids', then input_ids holds all input tokens. We can discard
-            # input_ids based on the past_length.
-            elif past_length < input_ids.shape[1]:
-                input_ids = input_ids[:, past_length:]
-            # 3 - Otherwise (past_length >= input_ids.shape[1]), let's assume input_ids only has unprocessed tokens.
-
-            # If the cache has seen more tokens than it can hold, then the cache has a size limit. Let's discard the
-            # older attention values, as their corresponding values are not part of the input.
-            if cache_length < past_length and attention_mask is not None:
-                attention_mask = attention_mask[
-                    :, -(cache_length + input_ids.shape[1]) :
-                ]
-
-        position_ids = kwargs.get("position_ids", None)
-        if attention_mask is not None and position_ids is None:
-            # create position_ids on the fly for batch generation
-            position_ids = attention_mask.long().cumsum(-1) - 1
-            position_ids.masked_fill_(attention_mask == 0, 1)
-            if past_key_values:
-                position_ids = position_ids[:, -input_ids.shape[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(
-            {
-                "position_ids": position_ids,
-                "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.to(past_state.device))
-                    for past_state in layer_past
-                ),
-            )
-        return reordered_past
-
-
-@add_start_docstrings(
-    """
-    The Mistral Model transformer with a sequence classification head on top (linear layer).
-
-    [`MistralForSequenceClassification`] uses the last token in order to do the classification, as other causal models
-    (e.g. GPT-2) 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).
-    """,
-    MISTRAL_START_DOCSTRING,
-)
-# Copied from transformers.models.llama.modeling_llama.LlamaForSequenceClassification with Llama->Mistral, LLAMA->MISTRAL
-class MistralForSequenceClassification(MixtralPreTrainedModel):
-    def __init__(self, config):
-        super().__init__(config)
-        self.num_labels = config.num_labels
-        self.model = MistralModel(config)
-        self.score = nn.Linear(config.hidden_size, self.num_labels, bias=False)
-
-        # Initialize weights and apply final processing
-        self.post_init()
-
-    def get_input_embeddings(self):
-        return self.model.embed_tokens
-
-    def set_input_embeddings(self, value):
-        self.model.embed_tokens = value
-
-    @add_start_docstrings_to_model_forward(MISTRAL_INPUTS_DOCSTRING)
-    def forward(
-        self,
-        input_ids: torch.LongTensor = None,
-        attention_mask: Optional[torch.Tensor] = None,
-        position_ids: Optional[torch.LongTensor] = 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, SequenceClassifierOutputWithPast]:
-        r"""
-        labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
-            Labels for computing the sequence classification/regression loss. Indices should be in `[0, ...,
-            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).
-        """
-        return_dict = (
-            return_dict if return_dict is not None else self.config.use_return_dict
-        )
-
-        transformer_outputs = self.model(
-            input_ids,
-            attention_mask=attention_mask,
-            position_ids=position_ids,
-            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,
-        )
-        hidden_states = transformer_outputs[0]
-        logits = self.score(hidden_states)
-
-        if input_ids is not None:
-            batch_size = input_ids.shape[0]
-        else:
-            batch_size = inputs_embeds.shape[0]
-
-        if self.config.pad_token_id is None and batch_size != 1:
-            raise ValueError(
-                "Cannot handle batch sizes > 1 if no padding token is defined."
-            )
-        if self.config.pad_token_id is None:
-            sequence_lengths = -1
-        else:
-            if input_ids is not None:
-                sequence_lengths = (
-                    torch.eq(input_ids, self.config.pad_token_id).int().argmax(-1) - 1
-                ).to(logits.device)
-            else:
-                sequence_lengths = -1
-
-        pooled_logits = logits[
-            torch.arange(batch_size, device=logits.device), sequence_lengths
-        ]
-
-        loss = None
-        if labels is not None:
-            labels = labels.to(logits.device)
-            if self.config.problem_type is None:
-                if self.num_labels == 1:
-                    self.config.problem_type = "regression"
-                elif self.num_labels > 1 and (
-                    labels.dtype == torch.long or labels.dtype == torch.int
-                ):
-                    self.config.problem_type = "single_label_classification"
-                else:
-                    self.config.problem_type = "multi_label_classification"
-
-            if self.config.problem_type == "regression":
-                loss_fct = MSELoss()
-                if self.num_labels == 1:
-                    loss = loss_fct(pooled_logits.squeeze(), labels.squeeze())
-                else:
-                    loss = loss_fct(pooled_logits, labels)
-            elif self.config.problem_type == "single_label_classification":
-                loss_fct = CrossEntropyLoss()
-                loss = loss_fct(
-                    pooled_logits.view(-1, self.num_labels), labels.view(-1)
-                )
-            elif self.config.problem_type == "multi_label_classification":
-                loss_fct = BCEWithLogitsLoss()
-                loss = loss_fct(pooled_logits, labels)
-        if not return_dict:
-            output = (pooled_logits,) + transformer_outputs[1:]
-            return ((loss,) + output) if loss is not None else output
-
-        return SequenceClassifierOutputWithPast(
-            loss=loss,
-            logits=pooled_logits,
-            past_key_values=transformer_outputs.past_key_values,
-            hidden_states=transformer_outputs.hidden_states,
-            attentions=transformer_outputs.attentions,
-        )

src/axolotl/monkeypatch/mixtral/__init__.py

@@ -0,0 +1,22 @@
+"""
+Patches to support multipack for mixtral
+"""
+import transformers
+
+
+def replace_mixtral_attn_with_multipack_flash_attn():
+    from .modeling_mixtral import (
+        MixtralMultipackFlashAttention2,
+        mixtral_decoder_layer_forward,
+        mixtral_model_forward,
+    )
+
+    transformers.models.mixtral.modeling_mixtral.MixtralDecoderLayer.forward = (
+        mixtral_decoder_layer_forward
+    )
+    transformers.models.mixtral.modeling_mixtral.MixtralModel.forward = (
+        mixtral_model_forward
+    )
+    transformers.models.mixtral.modeling_mixtral.MISTRAL_ATTENTION_CLASSES[
+        "flash_attention_2"
+    ] = MixtralMultipackFlashAttention2

src/axolotl/monkeypatch/mixtral/modeling_mixtral.py

@@ -0,0 +1,379 @@
+"""
+Mixtral modeling for multipack
+"""
+# pylint: disable=missing-module-docstring,unused-argument,protected-access,pointless-string-statement,duplicate-code
+import logging
+import warnings
+from typing import List, Optional, Tuple, Union
+
+import torch
+from einops import rearrange
+from flash_attn import flash_attn_varlen_qkvpacked_func
+from transformers import Cache, DynamicCache
+from transformers.modeling_attn_mask_utils import _prepare_4d_causal_attention_mask
+from transformers.modeling_outputs import MoeModelOutputWithPast
+from transformers.models.mixtral.modeling_mixtral import (
+    MixtralFlashAttention2,
+    apply_rotary_pos_emb,
+    repeat_kv,
+)
+
+from axolotl.monkeypatch.utils import get_cu_seqlens_from_pos_ids
+
+LOG = logging.getLogger("axolotl.monkeypatch.mixtral")
+
+
+class MixtralMultipackFlashAttention2(MixtralFlashAttention2):
+    """
+    Custom multipack implementation w flash attention 2
+    """
+
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+        self._flash_attn_uses_top_left_mask = True
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_value: Optional[Cache] = None,
+        output_attentions: bool = False,
+        use_cache: bool = False,
+        cu_seqlens: Optional[torch.Tensor] = None,
+        max_seqlen: Optional[torch.Tensor] = None,
+        **kwargs,
+    ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
+        if "padding_mask" in kwargs:
+            warnings.warn(
+                "Passing `padding_mask` is deprecated and will be removed in v4.37. Please make sure use `attention_mask` instead.`"
+            )
+        bsz, q_len, _ = hidden_states.size()
+
+        query_states = self.q_proj(hidden_states)
+        key_states = self.k_proj(hidden_states)
+        value_states = self.v_proj(hidden_states)
+
+        query_states = query_states.view(
+            bsz, q_len, self.num_heads, self.head_dim
+        ).transpose(1, 2)
+        key_states = key_states.view(
+            bsz, q_len, self.num_key_value_heads, self.head_dim
+        ).transpose(1, 2)
+        value_states = value_states.view(
+            bsz, q_len, self.num_key_value_heads, self.head_dim
+        ).transpose(1, 2)
+
+        kv_seq_len = key_states.shape[-2]
+        if past_key_value is not None:
+            if self.layer_idx is None:
+                raise ValueError(
+                    f"The cache structure has changed since version v4.36. If you are using {self.__class__.__name__} "
+                    "for auto-regressive decoding with k/v caching, please make sure to initialize the attention class "
+                    "with a layer index."
+                )
+            kv_seq_len += past_key_value.get_usable_length(kv_seq_len, self.layer_idx)
+        cos, sin = self.rotary_emb(value_states, seq_len=kv_seq_len)
+        query_states, key_states = apply_rotary_pos_emb(
+            query_states, key_states, cos, sin, position_ids
+        )
+
+        if past_key_value is not None:
+            cache_kwargs = {"sin": sin, "cos": cos}  # Specific to RoPE models
+            key_states, value_states = past_key_value.update(
+                key_states, value_states, self.layer_idx, cache_kwargs
+            )
+
+        # repeat k/v heads if n_kv_heads < n_heads
+        key_states = repeat_kv(key_states, self.num_key_value_groups)
+        value_states = repeat_kv(value_states, self.num_key_value_groups)
+
+        if cu_seqlens is not None and max_seqlen is not None and cu_seqlens.dim() == 1:
+            # special handling using sample packing
+            qkv = torch.stack(
+                [query_states, key_states, value_states], dim=2
+            )  # [bsz, nh, 3, q_len, hd]
+            qkv = qkv.transpose(1, 3)  # [bsz, q_len, 3, nh, hd]
+            qkv = rearrange(qkv, "b s ... -> (b s) ...")
+
+            attn_output = flash_attn_varlen_qkvpacked_func(
+                qkv,
+                cu_seqlens,
+                max_seqlen,
+                dropout_p=self.attention_dropout,
+                softmax_scale=None,
+                causal=True,
+            )
+            attn_output = rearrange(attn_output, "(b s) ... -> b s ...", b=bsz)
+
+        attn_output = attn_output.reshape(bsz, q_len, self.hidden_size).contiguous()
+        attn_output = self.o_proj(attn_output)
+
+        if not output_attentions:
+            attn_weights = None
+
+        return attn_output, attn_weights, past_key_value
+
+
+def mixtral_decoder_layer_forward(
+    self,
+    hidden_states: torch.Tensor,
+    attention_mask: Optional[torch.Tensor] = None,
+    position_ids: Optional[torch.LongTensor] = None,
+    past_key_value: Optional[Tuple[torch.Tensor]] = None,
+    output_attentions: Optional[bool] = False,
+    output_router_logits: Optional[bool] = False,
+    use_cache: Optional[bool] = False,
+    cu_seqlens: Optional[torch.Tensor] = None,
+    max_seqlen: Optional[torch.Tensor] = None,
+    **kwargs,
+) -> Tuple[torch.FloatTensor, Optional[Tuple[torch.FloatTensor, torch.FloatTensor]]]:
+    if "padding_mask" in kwargs:
+        warnings.warn(
+            "Passing `padding_mask` is deprecated and will be removed in v4.37. Please make sure use `attention_mask` instead.`"
+        )
+    """
+    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, sequence_length)` where padding elements are indicated by 0.
+        past_key_value (`Tuple(torch.FloatTensor)`, *optional*): cached past key and value projection states
+        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_router_logits (`bool`, *optional*):
+            Whether or not to return the logits of all the routers. They are useful for computing the router loss, and
+            should not be returned during inference.
+        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`).
+    """
+
+    residual = hidden_states
+
+    hidden_states = self.input_layernorm(hidden_states)
+
+    # Self Attention
+    hidden_states, self_attn_weights, present_key_value = self.self_attn(
+        hidden_states=hidden_states,
+        attention_mask=attention_mask,
+        position_ids=position_ids,
+        past_key_value=past_key_value,
+        output_attentions=output_attentions,
+        use_cache=use_cache,
+        cu_seqlens=cu_seqlens,
+        max_seqlen=max_seqlen,
+    )
+    hidden_states = residual + hidden_states
+
+    # Fully Connected
+    residual = hidden_states
+    hidden_states = self.post_attention_layernorm(hidden_states)
+    hidden_states, router_logits = self.block_sparse_moe(hidden_states)
+    hidden_states = residual + hidden_states
+
+    outputs = (hidden_states,)
+
+    if output_attentions:
+        outputs += (self_attn_weights,)
+
+    if use_cache:
+        outputs += (present_key_value,)
+
+    if output_router_logits:
+        outputs += (router_logits,)
+
+    return outputs
+
+
+def mixtral_model_forward(
+    self,
+    input_ids: torch.LongTensor = None,
+    attention_mask: Optional[torch.Tensor] = None,
+    position_ids: Optional[torch.LongTensor] = 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,
+    output_router_logits: Optional[bool] = None,
+    return_dict: Optional[bool] = None,
+) -> Union[Tuple, MoeModelOutputWithPast]:
+    output_attentions = (
+        output_attentions
+        if output_attentions is not None
+        else self.config.output_attentions
+    )
+    output_router_logits = (
+        output_router_logits
+        if output_router_logits is not None
+        else self.config.output_router_logits
+    )
+    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"
+        )
+    if input_ids is not None:
+        batch_size, seq_length = input_ids.shape
+    elif inputs_embeds is not None:
+        batch_size, seq_length, _ = inputs_embeds.shape
+    else:
+        raise ValueError(
+            "You have to specify either decoder_input_ids or decoder_inputs_embeds"
+        )
+
+    past_key_values_length = 0
+
+    if use_cache:
+        use_legacy_cache = not isinstance(past_key_values, Cache)
+        if use_legacy_cache:
+            past_key_values = DynamicCache.from_legacy_cache(past_key_values)
+        past_key_values_length = past_key_values.get_usable_length(seq_length)
+
+    cu_seqlens = None
+    max_seqlen = None
+    if position_ids is None:
+        device = input_ids.device if input_ids is not None else inputs_embeds.device
+        position_ids = torch.arange(
+            past_key_values_length,
+            seq_length + past_key_values_length,
+            dtype=torch.long,
+            device=device,
+        )
+        position_ids = position_ids.unsqueeze(0).view(-1, seq_length)
+    else:
+        position_ids = position_ids.view(-1, seq_length).long()
+        cu_seqlens, max_seqlen = get_cu_seqlens_from_pos_ids(position_ids)
+        cu_seqlens = cu_seqlens.squeeze()
+
+    if inputs_embeds is None:
+        inputs_embeds = self.embed_tokens(input_ids)
+
+    if attention_mask is not None and self._use_flash_attention_2 and use_cache:
+        is_padding_right = attention_mask[:, -1].sum().item() != batch_size
+        if is_padding_right:
+            raise ValueError(
+                "You are attempting to perform batched generation with padding_side='right'"
+                " this may lead to unexpected behaviour for Flash Attention version of Mixtral. Make sure to "
+                " call `tokenizer.padding_side  = 'left'` before tokenizing the input. "
+            )
+
+    if self._use_flash_attention_2:
+        # 2d mask is passed through the layers
+        attention_mask = (
+            attention_mask
+            if (attention_mask is not None and 0 in attention_mask)
+            else None
+        )
+    else:
+        # 4d mask is passed through the layers
+        attention_mask = _prepare_4d_causal_attention_mask(
+            attention_mask,
+            (batch_size, seq_length),
+            inputs_embeds,
+            past_key_values_length,
+            sliding_window=self.config.sliding_window,
+        )
+
+    hidden_states = inputs_embeds
+
+    if self.gradient_checkpointing and self.training:
+        if use_cache:
+            LOG.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
+    all_router_logits = () if output_router_logits else None
+    next_decoder_cache = None
+
+    for decoder_layer in self.layers:
+        if output_hidden_states:
+            all_hidden_states += (hidden_states,)
+
+        if self.gradient_checkpointing and self.training:
+            layer_outputs = self._gradient_checkpointing_func(
+                decoder_layer.__call__,
+                hidden_states,
+                attention_mask,
+                position_ids,
+                past_key_values,
+                output_attentions,
+                output_router_logits,
+                use_cache,
+                cu_seqlens,
+                max_seqlen,
+            )
+        else:
+            layer_outputs = decoder_layer(
+                hidden_states,
+                attention_mask=attention_mask,
+                position_ids=position_ids,
+                past_key_value=past_key_values,
+                output_attentions=output_attentions,
+                output_router_logits=output_router_logits,
+                use_cache=use_cache,
+                cu_seqlens=cu_seqlens,
+                max_seqlen=max_seqlen,
+            )
+
+        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],)
+
+        if output_router_logits:
+            all_router_logits += (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 = None
+    if use_cache:
+        next_cache = (
+            next_decoder_cache.to_legacy_cache()
+            if use_legacy_cache
+            else next_decoder_cache
+        )
+
+    if not return_dict:
+        return tuple(
+            v
+            for v in [
+                hidden_states,
+                next_cache,
+                all_hidden_states,
+                all_self_attns,
+                all_router_logits,
+            ]
+            if v is not None
+        )
+
+    return MoeModelOutputWithPast(
+        last_hidden_state=hidden_states,
+        past_key_values=next_cache,
+        hidden_states=all_hidden_states,
+        attentions=all_self_attns,
+        router_logits=all_router_logits,
+    )

src/axolotl/utils/models.py

@@ -54,25 +54,19 @@ def check_model_config(cfg: DictDefault, model_config: AutoConfig):
 def load_model_config(cfg):
     model_config_name = cfg.base_model_config or cfg.base_model
     trust_remote_code = cfg.trust_remote_code is True
-    model_type = cfg.model_type
-
-    if model_type == "MixtralForCausalLM":
-        from axolotl.models.mixtral.configuration_moe_mistral import MixtralConfig
 
-        model_config = MixtralConfig.from_pretrained(model_config_name)
-    else:
-        try:
-            model_config = AutoConfig.from_pretrained(
-                model_config_name, trust_remote_code=trust_remote_code
+    try:
+        model_config = AutoConfig.from_pretrained(
+            model_config_name, trust_remote_code=trust_remote_code
+        )
+    except ValueError as err:
+        if "mamba" in model_config_name:
+            return addict.Dict(
+                {
+                    "model_type": "mamba",
+                }
             )
-        except ValueError as err:
-            if "mamba" in model_config_name:
-                return addict.Dict(
-                    {
-                        "model_type": "mamba",
-                    }
-                )
-            raise err
+        raise err
 
     if cfg.model_config:
         for key, val in cfg.model_config.items():
@@ -255,6 +249,18 @@ def load_model(
         LOG.info("patching with flash attention")
         replace_mistral_attn_with_flash_attn(packed=cfg.sample_packing)
 
+    if (
+        cfg.model_config_type == "mixtral"
+        and cfg.flash_attention
+        and cfg.sample_packing
+    ):
+        from axolotl.monkeypatch.mixtral import (
+            replace_mixtral_attn_with_multipack_flash_attn,
+        )
+
+        LOG.info("patching with flash attention")
+        replace_mixtral_attn_with_multipack_flash_attn()
+
     if cfg.is_llama_derived_model and cfg.xpos_rope:
         from axolotl.monkeypatch.xpos_rope_llama_monkey_patch import (
             replace_llama_rope_with_xpos_rope,
@@ -302,15 +308,22 @@ def load_model(
             bnb_4bit_quant_type="nf4",
         )
     # sample packing uses custom FA2 patch
-    if cfg.flash_attention and not cfg.sample_packing:
-        if (
-            cfg.is_llama_derived_model
-            or cfg.is_falcon_derived_model
-            or cfg.is_mistral_derived_model
-        ):
-            # TODO enable once properly supported in transformers
-            # model_kwargs["attn_implementation"] = "flash_attention_2"
-            model_kwargs["use_flash_attention_2"] = True  # legacy, to be deprecated
+    if cfg.flash_attention:
+        if not cfg.sample_packing:
+            if (
+                cfg.is_llama_derived_model
+                or cfg.is_falcon_derived_model
+                or cfg.is_mistral_derived_model
+                or model_config.model_type == "mixtral"
+            ):
+                model_config._attn_implementation = (  # pylint: disable=protected-access
+                    "flash_attention_2"
+                )
+        else:
+            if model_config.model_type == "mixtral":
+                model_config._attn_implementation = (  # pylint: disable=protected-access
+                    "flash_attention_2"
+                )
 
     try:
         if cfg.is_llama_derived_model and not cfg.trust_remote_code and not cfg.gptq:
@@ -372,15 +385,6 @@ def load_model(
                 load_in_4bit=cfg.load_in_4bit and cfg.adapter is not None,
                 **model_kwargs,
             )
-        elif model_type == "MixtralForCausalLM":
-            from axolotl.models.mixtral import MixtralForCausalLM
-
-            model = MixtralForCausalLM.from_pretrained(
-                base_model,
-                load_in_8bit=cfg.load_in_8bit and cfg.adapter is not None,
-                load_in_4bit=cfg.load_in_4bit and cfg.adapter is not None,
-                **model_kwargs,
-            )
         elif model_type == "MambaLMHeadModel":
             # FIXME this is janky at best and hacked together to make it work
             MambaLMHeadModel = fix_mamba_attn_for_loss()  # pylint: disable=invalid-name