This repository contains minimal code to run our 7B, 8x7B and 8x22B models.

Blog 7B: https://mistral.ai/news/announcing-mistral-7b/
Blog 8x7B: https://mistral.ai/news/mixtral-of-experts/
Blog 8x22B: https://mistral.ai/news/mixtral-8x22b/
Blog Codestral 22B: https://mistral.ai/news/codestral
Blog Codestral Mamba 7B: https://mistral.ai/news/codestral-mamba/
Blog Mathstral 7B: https://mistral.ai/news/mathstral/
Blog Nemo: https://mistral.ai/news/mistral-nemo/
Blog Mistral Large 2: https://mistral.ai/news/mistral-large-2407/

Discord: https://discord.com/invite/mistralai
Documentation: https://docs.mistral.ai/
Guardrailing: https://docs.mistral.ai/usage/guardrailing

Note: You will use a GPU to install mistral-inference, as it currently requires xformers to be installed and xformers itself needs a GPU for installation.

pip install mistral-inference

cd $HOME && git clone https://github.com/mistralai/mistral-inference
cd $HOME/mistral-inference && poetry install .

• Important:
  • mixtral-8x22B-Instruct-v0.3.tar is exactly the same as Mixtral-8x22B-Instruct-v0.1, only stored in .safetensors format
  • mixtral-8x22B-v0.3.tar is the same as Mixtral-8x22B-v0.1, but has an extended vocabulary of 32768 tokens.
  • codestral-22B-v0.1.tar has a custom non-commercial license, called Mistral AI Non-Production (MNPL) License
  • mistral-large-instruct-2407.tar has a custom non-commercial license, called Mistral AI Research (MRL) License
• All of the listed models above support function calling. For example, Mistral 7B Base/Instruct v3 is a minor update to Mistral 7B Base/Instruct v2, with the addition of function calling capabilities.
• The "coming soon" models will include function calling as well.
• You can download the previous versions of our models from our docs.

News!!!: Mistral Large 2 is out. Read more about its capabilities here.

Create a local folder to store models

export MISTRAL_MODEL=$HOME/mistral_models
mkdir -p $MISTRAL_MODEL

Download any of the above links and extract the content, e.g.:

export M7B_DIR=$MISTRAL_MODEL/12B_Nemo
wget https://models.mistralcdn.com/mistral-nemo-2407/mistral-nemo-instruct-2407.tar
mkdir -p $12B_DIR
tar -xf mistral-nemo-instruct-v0.1.tar -C $12B_DIR

or

export M8x7B_DIR=$MISTRAL_MODEL/8x7b_instruct
wget https://models.mistralcdn.com/mixtral-8x7b-v0-1/Mixtral-8x7B-v0.1-Instruct.tar
mkdir -p $M8x7B_DIR
tar -xf Mixtral-8x7B-v0.1-Instruct.tar -C $M8x7B_DIR

The following sections give an overview of how to run the model from the Command-line interface (CLI) or directly within Python.

• Demo

To test that a model works in your setup, you can run the mistral-demo command. E.g. the 12B Mistral-Nemo model can be tested on a single GPU as follows:

mistral-demo $12B_DIR

Large models, such 8x7B and 8x22B have to be run in a multi-GPU setup. For these models, you can use the following command:

torchrun --nproc-per-node 2 --no-python mistral-demo $M8x7B_DIR

Note: Change --nproc-per-node to more GPUs if available.

• Chat

To interactively chat with the models, you can make use of the mistral-chat command.

mistral-chat $12B_DIR --instruct --max_tokens 1024 --temperature 0.35

For large models, you can make use of torchrun.

torchrun --nproc-per-node 2 --no-python mistral-chat $M8x7B_DIR --instruct

Note: Change --nproc-per-node to more GPUs if necessary (e.g. for 8x22B).

• Chat with Codestral

To use Codestral as a coding assistant you can run the following command using mistral-chat. Make sure $M22B_CODESTRAL is set to a valid path to the downloaded codestral folder, e.g. $HOME/mistral_models/Codestral-22B-v0.1

mistral-chat $M22B_CODESTRAL --instruct --max_tokens 256

If you prompt it with "Write me a function that computes fibonacci in Rust", the model should generate something along the following lines:

Sure, here's a simple implementation of a function that computes the Fibonacci sequence in Rust. This function takes an integer `n` as an argument and returns the `n`th Fibonacci number.

fn fibonacci(n: u32) -> u32 {
    match n {
        0 => 0,
        1 => 1,
        _ => fibonacci(n - 1) + fibonacci(n - 2),
    }
}

fn main() {
    let n = 10;
    println!("The {}th Fibonacci number is: {}", n, fibonacci(n));
}

This function uses recursion to calculate the Fibonacci number. However, it's not the most efficient solution because it performs a lot of redundant calculations. A more efficient solution would use a loop to iteratively calculate the Fibonacci numbers.

You can continue chatting afterwards, e.g. with "Translate it to Python".

• Chat with Codestral-Mamba

To use Codestral-Mamba as a coding assistant you can run the following command using mistral-chat. Make sure $7B_CODESTRAL_MAMBA is set to a valid path to the downloaded codestral-mamba folder, e.g. $HOME/mistral_models/mamba-codestral-7B-v0.1.

You then need to additionally install the following packages:

pip install packaging mamba-ssm causal-conv1d transformers

before you can start chatting:

mistral-chat $7B_CODESTRAL_MAMBA --instruct --max_tokens 256

• Chat with Mathstral

To use Mathstral as an assistant you can run the following command using mistral-chat. Make sure $7B_MATHSTRAL is set to a valid path to the downloaded codestral folder, e.g. $HOME/mistral_models/mathstral-7B-v0.1

mistral-chat $7B_MATHSTRAL --instruct --max_tokens 256

If you prompt it with "Albert likes to surf every week. Each surfing session lasts for 4 hours and costs $20 per hour. How much would Albert spend in 5 weeks?", the model should answer with the correct calculation.

You can then continue chatting afterwards, e.g. with "How much would he spend in a year?".

• Instruction Following:

from mistral_inference.transformer import Transformer
from mistral_inference.generate import generate

from mistral_common.tokens.tokenizers.mistral import MistralTokenizer
from mistral_common.protocol.instruct.messages import UserMessage
from mistral_common.protocol.instruct.request import ChatCompletionRequest


tokenizer = MistralTokenizer.from_file("./mistral-nemo-instruct-v0.1/tekken.json")  # change to extracted tokenizer file
model = Transformer.from_folder("./mistral-nemo-instruct-v0.1")  # change to extracted model dir

prompt = "How expensive would it be to ask a window cleaner to clean all windows in Paris. Make a reasonable guess in US Dollar."

completion_request = ChatCompletionRequest(messages=[UserMessage(content=prompt)])

tokens = tokenizer.encode_chat_completion(completion_request).tokens

out_tokens, _ = generate([tokens], model, max_tokens=1024, temperature=0.35, eos_id=tokenizer.instruct_tokenizer.tokenizer.eos_id)
result = tokenizer.instruct_tokenizer.tokenizer.decode(out_tokens[0])

print(result)

• Function Calling:

from mistral_common.protocol.instruct.tool_calls import Function, Tool

completion_request = ChatCompletionRequest(
    tools=[
        Tool(
            function=Function(
                name="get_current_weather",
                description="Get the current weather",
                parameters={
                    "type": "object",
                    "properties": {
                        "location": {
                            "type": "string",
                            "description": "The city and state, e.g. San Francisco, CA",
                        },
                        "format": {
                            "type": "string",
                            "enum": ["celsius", "fahrenheit"],
                            "description": "The temperature unit to use. Infer this from the users location.",
                        },
                    },
                    "required": ["location", "format"],
                },
            )
        )
    ],
    messages=[
        UserMessage(content="What's the weather like today in Paris?"),
        ],
)

tokens = tokenizer.encode_chat_completion(completion_request).tokens

out_tokens, _ = generate([tokens], model, max_tokens=64, temperature=0.0, eos_id=tokenizer.instruct_tokenizer.tokenizer.eos_id)
result = tokenizer.instruct_tokenizer.tokenizer.decode(out_tokens[0])

print(result)

• Fill-in-the-middle (FIM):

Make sure to have mistral-common >= 1.2.0 installed:

pip install --upgrade mistral-common

You can simulate a code completion in-filling as follows.

from mistral_inference.transformer import Transformer
from mistral_inference.generate import generate
from mistral_common.tokens.tokenizers.mistral import MistralTokenizer
from mistral_common.tokens.instruct.request import FIMRequest

tokenizer = MistralTokenizer.from_model("codestral-22b")
model = Transformer.from_folder("./mistral_22b_codestral")

prefix = """def add("""
suffix = """    return sum"""

request = FIMRequest(prompt=prefix, suffix=suffix)

tokens = tokenizer.encode_fim(request).tokens

out_tokens, _ = generate([tokens], model, max_tokens=256, temperature=0.0, eos_id=tokenizer.instruct_tokenizer.tokenizer.eos_id)
result = tokenizer.decode(out_tokens[0])

middle = result.split(suffix)[0].strip()
print(middle)

If you want a self-contained implementation, look at one_file_ref.py, or run it with

python -m one_file_ref $M7B_DIR

which should give something along the following lines:

This is a test of the emergency broadcast system. This is only a test.

If this were a real emergency, you would be told what to do.

This is a test
=====================
This is another test of the new blogging software. I’m not sure if I’m going to keep it or not. I’m not sure if I’m going to keep
=====================
This is a third test, mistral AI is very good at testing. 🙂

This is a third test, mistral AI is very good at testing. 🙂

This
=====================

Note: To run self-contained implementations, you need to do a local installation.

To run logits equivalence:

python -m pytest tests

The deploy folder contains code to build a vLLM image with the required dependencies to serve the Mistral AI model. In the image, the transformers library is used instead of the reference implementation. To build it:

docker build deploy --build-arg MAX_JOBS=8

Instructions to run the image can be found in the official documentation.

• Use Mistral models on Mistral AI official API (La Plateforme)
• Use Mistral models via cloud providers

[1]: LoRA: Low-Rank Adaptation of Large Language Models, Hu et al. 2021