Mixtral-8x7B is Mistral AI’s second Large Language Model (LLM).
The Mixtral model was proposed by the Mistral AI team.
It was introduced in the Mixtral of Experts blogpost with the following introduction:
Today, the team is proud to release Mixtral 8x7B, a high-quality sparse mixture of experts models (SMoE) with open weights. Licensed under Apache 2.0. Mixtral outperforms Llama 2 70B on most benchmarks with 6x faster inference. It is the strongest open-weight model with a permissive license and the best model overall regarding cost/performance trade-offs. In particular, it matches or outperforms GPT3.5 on most standard benchmarks.
Tips:
This model was contributed by Younes Belkada and Arthur Zucker . The original code can be found here.
Mixtral-45B is a decoder-based LM with the following architectural choices:
The following implementation details are shared with Mistral AI’s first model mistral:
They also provide an instruction fine-tuned model: mistralai/Mixtral-8x7B-v0.1 which can be used for chat-based inference.
For more details please read our release blog post
Mixtral-8x7B is released under the Apache 2.0 license.
Mixtral-8x7B can be found on the Huggingface Hub
These ready-to-use checkpoints can be downloaded and used via the HuggingFace Hub:
>>> from transformers import AutoModelForCausalLM, AutoTokenizer
>>> device = "cuda" # the device to load the model onto
>>> model = AutoModelForCausalLM.from_pretrained("mistralai/Mixtral-8x7B-v0.1")
>>> tokenizer = AutoTokenizer.from_pretrained("mistralai/Mixtral-8x7B-v0.1")
>>> prompt = "My favourite condiment is"
>>> model_inputs = tokenizer([prompt], return_tensors="pt").to(device)
>>> model.to(device)
>>> generated_ids = model.generate(**model_inputs, max_new_tokens=100, do_sample=True)
>>> tokenizer.batch_decode(generated_ids)[0]
"The expected output"To use the raw checkpoints with HuggingFace you can use the convert_mixtral_weights_to_hf.py script to convert them to the HuggingFace format:
python src/transformers/models/mixtral/convert_mixtral_weights_to_hf.py \
--input_dir /path/to/downloaded/mistral/weights --output_dir /output/pathYou can then load the converted model from the output/path:
from transformers import MixtralForCausalLM, LlamaTokenizer
tokenizer = LlamaTokenizer.from_pretrained("/output/path")
model = MixtralForCausalLM.from_pretrained("/output/path")First, make sure to install the latest version of Flash Attention 2 to include the sliding window attention feature.
pip install -U flash-attn --no-build-isolation
Make also sure that you have a hardware that is compatible with Flash-Attention 2. Read more about it in the official documentation of flash-attn repository. Make also sure to load your model in half-precision (e.g. torch.float16)
To load and run a model using Flash Attention 2, refer to the snippet below:
>>> import torch
>>> from transformers import AutoModelForCausalLM, AutoTokenizer
>>> device = "cuda" # the device to load the model onto
>>> model = AutoModelForCausalLM.from_pretrained("mistralai/Mixtral-8x7B-v0.1", torch_dtype=torch.float16, attn_implementation="flash_attention_2")
>>> tokenizer = AutoTokenizer.from_pretrained("mistralai/Mixtral-8x7B-v0.1")
>>> prompt = "My favourite condiment is"
>>> model_inputs = tokenizer([prompt], return_tensors="pt").to(device)
>>> model.to(device)
>>> generated_ids = model.generate(**model_inputs, max_new_tokens=100, do_sample=True)
>>> tokenizer.batch_decode(generated_ids)[0]
"The expected output"Below is a expected speedup diagram that compares pure inference time between the native implementation in transformers using mistralai/Mixtral-8x7B-v0.1 checkpoint and the Flash Attention 2 version of the model.
The current implementation supports the sliding window attention mechanism and memory efficient cache management.
To enable sliding window attention, just make sure to have a flash-attn version that is compatible with sliding window attention (>=2.3.0).
The Flash Attention-2 model uses also a more memory efficient cache slicing mechanism - as recommended per the official implementation of Mistral model that use rolling cache mechanism we keep the cache size fixed (self.config.sliding_window), support batched generation only for padding_side="left" and use the absolute position of the current token to compute the positional embedding.
Albert Jiang, Alexandre Sablayrolles, Arthur Mensch, Chris Bamford, Devendra Singh Chaplot, Diego de las Casas, Florian Bressand, Gianna Lengyel, Guillaume Lample, Lélio Renard Lavaud, Lucile Saulnier, Marie-Anne Lachaux, Pierre Stock, Teven Le Scao, Thibaut Lavril, Thomas Wang, Timothée Lacroix, William El Sayed.
( 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 = 131072 initializer_range = 0.02 rms_norm_eps = 1e-05 use_cache = True pad_token_id = None bos_token_id = 1 eos_token_id = 2 tie_word_embeddings = False rope_theta = 1000000.0 sliding_window = None attention_dropout = 0.0 num_experts_per_tok = 2 num_local_experts = 8 output_router_logits = False router_aux_loss_coef = 0.001 **kwargs )
Parameters
int, optional, defaults to 32000) —
Vocabulary size of the Mixtral model. Defines the number of different tokens that can be represented by the
inputs_ids passed when calling MixtralModel int, optional, defaults to 4096) —
Dimension of the hidden representations. int, optional, defaults to 14336) —
Dimension of the MLP representations. int, optional, defaults to 32) —
Number of hidden layers in the Transformer encoder. int, optional, defaults to 32) —
Number of attention heads for each attention layer in the Transformer encoder. 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`. str or function, optional, defaults to "silu") —
The non-linear activation function (function or string) in the decoder. int, optional, defaults to 4096*32) —
The maximum sequence length that this model might ever be used with. Mixtral’s sliding window attention
allows sequence of up to 4096*32 tokens. float, optional, defaults to 0.02) —
The standard deviation of the truncated_normal_initializer for initializing all weight matrices. float, optional, defaults to 1e-05) —
The epsilon used by the rms normalization layers. 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. int, optional) —
The id of the padding token. int, optional, defaults to 1) —
The id of the “beginning-of-sequence” token. int, optional, defaults to 2) —
The id of the “end-of-sequence” token. bool, optional, defaults to False) —
Whether the model’s input and output word embeddings should be tied. float, optional, defaults to 1000000.0) —
The base period of the RoPE embeddings. int, optional) —
Sliding window attention window size. If not specified, will default to 4096. float, optional, defaults to 0.0) —
The dropout ratio for the attention probabilities. int, optional, defaults to 2) —
The number of experts to root per-token, can be also interpreted as the top-p routing
parameter int, optional, defaults to 8) —
Number of experts per Sparse MLP layer. bool, optional, defaults to False) —
Whether or not the router logits should be returned by the model. Enabeling this will also
allow the model to output the auxiliary loss. See here for more details float, optional, defaults to 0.001) —
The aux loss factor for the total loss. This is the configuration class to store the configuration of a MixtralModel. It is used to instantiate an Mixtral 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 Mixtral-7B-v0.1 or Mixtral-7B-Instruct-v0.1.
mixtralai/Mixtral-8x7B mixtralai/Mixtral-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.
>>> from transformers import MixtralModel, MixtralConfig
>>> # Initializing a Mixtral 7B style configuration
>>> configuration = MixtralConfig()
>>> # Initializing a model from the Mixtral 7B style configuration
>>> model = MixtralModel(configuration)
>>> # Accessing the model configuration
>>> configuration = model.config( config: MixtralConfig )
Parameters
The bare Mixtral Model outputting raw hidden-states without any specific head on top. 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 subclass. Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and behavior.
Transformer decoder consisting of config.num_hidden_layers layers. Each layer is a MixtralDecoderLayer
( input_ids: LongTensor = None attention_mask: Optional = None position_ids: Optional = None past_key_values: Optional = None inputs_embeds: Optional = None use_cache: Optional = None output_attentions: Optional = None output_hidden_states: Optional = None output_router_logits: Optional = None return_dict: Optional = None )
Parameters
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.
torch.Tensor of shape (batch_size, sequence_length), optional) —
Mask to avoid performing attention on padding token indices. Mask values selected in [0, 1]:
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 for more
information on the default strategy.
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].
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).
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. 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). bool, optional) —
Whether or not to return the attentions tensors of all attention layers. See attentions under returned
tensors for more detail. bool, optional) —
Whether or not to return the hidden states of all layers. See hidden_states under returned tensors for
more detail. 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. bool, optional) —
Whether or not to return a ModelOutput instead of a plain tuple. The MixtralModel forward method, overrides the __call__ special method.
Although the recipe for forward pass needs to be defined within this function, one should call the Module
instance afterwards instead of this since the former takes care of running the pre and post processing steps while
the latter silently ignores them.
( input_ids: LongTensor = None attention_mask: Optional = None position_ids: Optional = None past_key_values: Optional = None inputs_embeds: Optional = None labels: Optional = None use_cache: Optional = None output_attentions: Optional = None output_hidden_states: Optional = None output_router_logits: Optional = None return_dict: Optional = None ) → transformers.modeling_outputs.MoeCausalLMOutputWithPast or tuple(torch.FloatTensor)
Parameters
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.
torch.Tensor of shape (batch_size, sequence_length), optional) —
Mask to avoid performing attention on padding token indices. Mask values selected in [0, 1]:
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 for more
information on the default strategy.
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].
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).
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. 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). bool, optional) —
Whether or not to return the attentions tensors of all attention layers. See attentions under returned
tensors for more detail. bool, optional) —
Whether or not to return the hidden states of all layers. See hidden_states under returned tensors for
more detail. 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. bool, optional) —
Whether or not to return a ModelOutput instead of a plain tuple.
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
transformers.modeling_outputs.MoeCausalLMOutputWithPast or tuple(torch.FloatTensor)
A transformers.modeling_outputs.MoeCausalLMOutputWithPast or a tuple of
torch.FloatTensor (if return_dict=False is passed or when config.return_dict=False) comprising various
elements depending on the configuration (MixtralConfig) and inputs.
loss (torch.FloatTensor of shape (1,), optional, returned when labels is provided) — Language modeling loss (for next-token prediction).
logits (torch.FloatTensor of shape (batch_size, sequence_length, config.vocab_size)) — Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).
aux_loss (torch.FloatTensor, optional, returned when labels is provided) — aux_loss for the sparse modules.
router_logits (tuple(torch.FloatTensor), optional, returned when output_router_probs=True and config.add_router_probs=True is passed or when config.output_router_probs=True) — Tuple of torch.FloatTensor (one for each layer) of shape (batch_size, sequence_length, num_experts).
Raw router logtis (post-softmax) that are computed by MoE routers, these terms are used to compute the auxiliary loss for Mixture of Experts models.
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))
Contains pre-computed hidden-states (key and values in the self-attention blocks) that can be used (see
past_key_values input) to speed up sequential decoding.
hidden_states (tuple(torch.FloatTensor), optional, returned when output_hidden_states=True is passed or when config.output_hidden_states=True) — Tuple of torch.FloatTensor (one for the output of the embeddings, if the model has an embedding layer, +
one for the output of each layer) of shape (batch_size, sequence_length, hidden_size).
Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.
attentions (tuple(torch.FloatTensor), optional, returned when output_attentions=True is passed or when config.output_attentions=True) — Tuple of torch.FloatTensor (one for each layer) of shape (batch_size, num_heads, sequence_length, sequence_length).
Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.
The MixtralForCausalLM forward method, overrides the __call__ special method.
Although the recipe for forward pass needs to be defined within this function, one should call the Module
instance afterwards instead of this since the former takes care of running the pre and post processing steps while
the latter silently ignores them.
Example:
>>> from transformers import AutoTokenizer, MixtralForCausalLM
>>> model = MixtralForCausalLM.from_pretrained("mistralai/Mixtral-8x7B-v0.1")
>>> tokenizer = AutoTokenizer.from_pretrained("mistralai/Mixtral-8x7B-v0.1")
>>> 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."( config )
Parameters
The Mixtral Model transformer with a sequence classification head on top (linear layer).
MixtralForSequenceClassification 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).
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 subclass. Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and behavior.
( input_ids: LongTensor = None attention_mask: Optional = None position_ids: Optional = None past_key_values: Optional = None inputs_embeds: Optional = None labels: Optional = None use_cache: Optional = None output_attentions: Optional = None output_hidden_states: Optional = None return_dict: Optional = None )
Parameters
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.
torch.Tensor of shape (batch_size, sequence_length), optional) —
Mask to avoid performing attention on padding token indices. Mask values selected in [0, 1]:
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 for more
information on the default strategy.
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].
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).
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. 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). bool, optional) —
Whether or not to return the attentions tensors of all attention layers. See attentions under returned
tensors for more detail. bool, optional) —
Whether or not to return the hidden states of all layers. See hidden_states under returned tensors for
more detail. 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. bool, optional) —
Whether or not to return a ModelOutput instead of a plain tuple. 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). The MixtralForSequenceClassification forward method, overrides the __call__ special method.
Although the recipe for forward pass needs to be defined within this function, one should call the Module
instance afterwards instead of this since the former takes care of running the pre and post processing steps while
the latter silently ignores them.