diff --git a/docs/source/en/model_doc/mistral.md b/docs/source/en/model_doc/mistral.md index 097d8888f9a5..39334385e605 100644 --- a/docs/source/en/model_doc/mistral.md +++ b/docs/source/en/model_doc/mistral.md @@ -14,74 +14,55 @@ rendered properly in your Markdown viewer. --> -# Mistral - -
-PyTorch -TensorFlow -Flax -FlashAttention -SDPA +
+
+ PyTorch + TensorFlow + Flax + FlashAttention + SDPA +
-## Overview - -Mistral was introduced in the [this blogpost](https://mistral.ai/news/announcing-mistral-7b/) by 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. - -The introduction of the blog post says: - -*Mistral AI team is proud to release Mistral 7B, the most powerful language model for its size to date.* - -Mistral-7B is the first large language model (LLM) released by [mistral.ai](https://mistral.ai/). - -### Architectural details - -Mistral-7B is a decoder-only Transformer with the following architectural choices: - -- Sliding Window Attention - Trained with 8k context length and fixed cache size, with a theoretical attention span of 128K tokens -- GQA (Grouped Query Attention) - allowing faster inference and lower cache size. -- Byte-fallback BPE tokenizer - ensures that characters are never mapped to out of vocabulary tokens. - -For more details refer to the [release blog post](https://mistral.ai/news/announcing-mistral-7b/). - -### License +# Mistral -`Mistral-7B` is released under the Apache 2.0 license. +[Mistral](https://huggingface.co/papers/2310.06825) is a 7B parameter language model, available as a pretrained and instruction-tuned variant, focused on balancing +the scaling costs of large models with performance and efficient inference. This model uses sliding window attention (SWA) trained with a 8K context length and a fixed cache size to handle longer sequences more effectively. Grouped-query attention (GQA) speeds up inference and reduces memory requirements. Mistral also features a byte-fallback BPE tokenizer to improve token handling and efficiency by ensuring characters are never mapped to out-of-vocabulary tokens. -## Usage tips +You can find all the original Mistral checkpoints under the [Mistral AI_](https://huggingface.co/mistralai) organization. -The Mistral team has released 3 checkpoints: +> [!TIP] +> Click on the Mistral models in the right sidebar for more examples of how to apply Mistral to different language tasks. -- a base model, [Mistral-7B-v0.1](https://huggingface.co/mistralai/Mistral-7B-v0.1), which has been pre-trained to predict the next token on internet-scale data. -- an instruction tuned model, [Mistral-7B-Instruct-v0.1](https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.1), which is the base model optimized for chat purposes using supervised fine-tuning (SFT) and direct preference optimization (DPO). -- an improved instruction tuned model, [Mistral-7B-Instruct-v0.2](https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.2), which improves upon v1. +The example below demonstrates how to chat with [`Pipeline`] or the [`AutoModel`], and from the command line. -The base model can be used as follows: + + ```python ->>> from transformers import AutoModelForCausalLM, AutoTokenizer - ->>> model = AutoModelForCausalLM.from_pretrained("mistralai/Mistral-7B-v0.1", device_map="auto") ->>> tokenizer = AutoTokenizer.from_pretrained("mistralai/Mistral-7B-v0.1") - ->>> prompt = "My favourite condiment is" +>>> import torch +>>> from transformers import pipeline ->>> model_inputs = tokenizer([prompt], return_tensors="pt").to("cuda") ->>> model.to(device) +>>> messages = [ +... {"role": "user", "content": "What is your favourite condiment?"}, +... {"role": "assistant", "content": "Well, I'm quite partial to a good squeeze of fresh lemon juice. It adds just the right amount of zesty flavour to whatever I'm cooking up in the kitchen!"}, +... {"role": "user", "content": "Do you have mayonnaise recipes?"} +... ] ->>> generated_ids = model.generate(**model_inputs, max_new_tokens=100, do_sample=True) ->>> tokenizer.batch_decode(generated_ids)[0] -"My favourite condiment is to ..." +>>> chatbot = pipeline("text-generation", model="mistralai/Mistral-7B-Instruct-v0.3", torch_dtype=torch.bfloat16, device=0) +>>> chatbot(messages) ``` -The instruction tuned model can be used as follows: + + ```python +>>> import torch >>> from transformers import AutoModelForCausalLM, AutoTokenizer ->>> model = AutoModelForCausalLM.from_pretrained("mistralai/Mistral-7B-Instruct-v0.2", device_map="auto") ->>> tokenizer = AutoTokenizer.from_pretrained("mistralai/Mistral-7B-Instruct-v0.2") +>>> model = AutoModelForCausalLM.from_pretrained("mistralai/Mistral-7B-Instruct-v0.3", torch_dtype=torch.bfloat16, attn_implementation="sdpa", device_map="auto") +>>> tokenizer = AutoTokenizer.from_pretrained("mistralai/Mistral-7B-Instruct-v0.3") >>> messages = [ ... {"role": "user", "content": "What is your favourite condiment?"}, @@ -96,59 +77,20 @@ The instruction tuned model can be used as follows: "Mayonnaise can be made as follows: (...)" ``` -As can be seen, the instruction-tuned model requires a [chat template](../chat_templating) to be applied to make sure the inputs are prepared in the right format. - -## Speeding up Mistral by using Flash Attention - -The code snippets above showcase inference without any optimization tricks. However, one can drastically speed up the model by leveraging [Flash Attention](../perf_train_gpu_one#flash-attention-2), which is a faster implementation of the attention mechanism used inside the model. - -First, make sure to install the latest version of Flash Attention 2 to include the sliding window attention feature. - -```bash -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 the [flash attention repository](https://github.com/Dao-AILab/flash-attention). 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: + + ```python ->>> import torch ->>> from transformers import AutoModelForCausalLM, AutoTokenizer - ->>> model = AutoModelForCausalLM.from_pretrained("mistralai/Mistral-7B-v0.1", torch_dtype=torch.float16, attn_implementation="flash_attention_2", device_map="auto") ->>> tokenizer = AutoTokenizer.from_pretrained("mistralai/Mistral-7B-v0.1") - ->>> prompt = "My favourite condiment is" - ->>> model_inputs = tokenizer([prompt], return_tensors="pt").to("cuda") ->>> model.to(device) - ->>> generated_ids = model.generate(**model_inputs, max_new_tokens=100, do_sample=True) ->>> tokenizer.batch_decode(generated_ids)[0] -"My favourite condiment is to (...)" +echo -e "My favorite condiment is" | transformers-cli chat --model_name_or_path mistralai/Mistral-7B-v0.3 --torch_dtype auto --device 0 --attn_implementation flash_attention_2 ``` -### Expected speedups - -Below is a expected speedup diagram that compares pure inference time between the native implementation in transformers using `mistralai/Mistral-7B-v0.1` checkpoint and the Flash Attention 2 version of the model. - -
- -
- -### Sliding window Attention + + -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. +Quantization reduces the memory burden of large models by representing the weights in a lower precision. Refer to the [Quantization](../quantization/overview) overview for more available quantization backends. -## Shrinking down Mistral using quantization - -As the Mistral model has 7 billion parameters, that would require about 14GB of GPU RAM in half precision (float16), since each parameter is stored in 2 bytes. However, one can shrink down the size of the model using [quantization](../quantization.md). If the model is quantized to 4 bits (or half a byte per parameter),that requires only about 3.5GB of RAM. - -Quantizing a model is as simple as passing a `quantization_config` to the model. Below, we'll leverage the BitsAndyBytes quantization (but refer to [this page](../quantization.md) for other quantization methods): +The example below uses [bitsandbytes](../quantization/bitsandbytes) to only quantize the weights to 4-bits. ```python >>> import torch @@ -161,8 +103,8 @@ Quantizing a model is as simple as passing a `quantization_config` to the model. ... bnb_4bit_compute_dtype="torch.float16", ... ) ->>> model = AutoModelForCausalLM.from_pretrained("mistralai/Mistral-7B-Instruct-v0.2", quantization_config=True, device_map="auto") ->>> tokenizer = AutoTokenizer.from_pretrained("mistralai/Mistral-7B-Instruct-v0.2") +>>> model = AutoModelForCausalLM.from_pretrained("mistralai/Mistral-7B-Instruct-v0.3", quantization_config=True, torch_dtype=torch.bfloat16, device_map="auto") +>>> tokenizer = AutoTokenizer.from_pretrained("mistralai/Mistral-7B-Instruct-v0.3") >>> prompt = "My favourite condiment is" @@ -179,19 +121,18 @@ Quantizing a model is as simple as passing a `quantization_config` to the model. "The expected output" ``` -This model was contributed by [Younes Belkada](https://huggingface.co/ybelkada) and [Arthur Zucker](https://huggingface.co/ArthurZ) . -The original code can be found [here](https://github.com/mistralai/mistral-src). - -## Resources +Use the [AttentionMaskVisualizer](https://github.com/huggingface/transformers/blob/beb9b5b02246b9b7ee81ddf938f93f44cfeaad19/src/transformers/utils/attention_visualizer.py#L139) to better understand what tokens the model can and cannot attend to. -A list of official Hugging Face and community (indicated by 🌎) resources to help you get started with Mistral. If you're interested in submitting a resource to be included here, please feel free to open a Pull Request and we'll review it! The resource should ideally demonstrate something new instead of duplicating an existing resource. +```py +>>> from transformers.utils.attention_visualizer import AttentionMaskVisualizer - +>>> visualizer = AttentionMaskVisualizer("mistralai/Mistral-7B-Instruct-v0.3") +>>> visualizer("Do you have mayonnaise recipes?") +``` -- A demo notebook to perform supervised fine-tuning (SFT) of Mistral-7B can be found [here](https://github.com/NielsRogge/Transformers-Tutorials/blob/master/Mistral/Supervised_fine_tuning_(SFT)_of_an_LLM_using_Hugging_Face_tooling.ipynb). 🌎 -- A [blog post](https://www.philschmid.de/fine-tune-llms-in-2024-with-trl) on how to fine-tune LLMs in 2024 using Hugging Face tooling. 🌎 -- The [Alignment Handbook](https://github.com/huggingface/alignment-handbook) by Hugging Face includes scripts and recipes to perform supervised fine-tuning (SFT) and direct preference optimization with Mistral-7B. This includes scripts for full fine-tuning, QLoRa on a single GPU as well as multi-GPU fine-tuning. -- [Causal language modeling task guide](../tasks/language_modeling) +
+ +
## MistralConfig @@ -245,4 +186,4 @@ A list of official Hugging Face and community (indicated by 🌎) resources to h ## TFMistralForSequenceClassification [[autodoc]] TFMistralForSequenceClassification - - call \ No newline at end of file + - call