Mistral-7B-Instruct-v0.3

Mistral-7B-Instruct-v0.3

A 7.2B open chat model from Mistral AI that follows instructions and can call functions you describe. It is fine-tuned from Mistral-7B-v0.3 and runs with mistral_inference or transformers.

Made by mistralai on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: Mistral-7B-Instruct-v0.3 is a language model with 7.2B parameters, tuned to follow instructions. It starts from Mistral-7B-v0.3. Compared with v0.2, the underlying model has a vocabulary extended to 32768, works with the v3 tokenizer and supports function calling. You give it a chat request and it writes the reply, or a call to a function you described.

Who it suits: It suits developers who want a compact open chat model that can call functions, and who can run a 7B model on their own hardware.

Industry | Who uses it | How they use it | Example
Software development | App developers | Add a self-hosted chat assistant | Ask it for a short explanation of machine learning through mistral-chat
Automation | Teams that connect models to tools | Let the model call a function you define | Describe a weather function and ask about the weather in Paris
Research | Tinkerers and researchers | Compare compact instruction models | Start from this model and test your own prompts

If you need a larger model for hard reasoning, you can skip it.

Where to use it:

• A computer with enough memory for the weights, about 14.5 GB in BF16 or about 3.6 GB at 4-bit, which is an estimate
• Python scripts that use mistral_inference or transformers
• Apps that serve models with vLLM, the library the Hugging Face page lists
• Windows, macOS or Linux with Python, to download the files

Use case | Who | What they get back
Chat | Developers | Replies to the instructions you type
Function calling | Tool builders | A structured call to the function you describe
Persona replies | Hobbyists | Answers in a style you set with a system message, as in the pirate example

How to set it up: What you need first:

• Python
• mistral_inference, or transformers if you prefer it
• For function calling through transformers, version 4.42.0 or higher
• Room for the weights, about 14.5 GB in BF16

1. Install mistral_inference.
2. Download the three files the card lists into a folder in your home directory.
3. Run mistral-chat on that folder.

pip install mistral_inference

from huggingface_hub import snapshot_download
from pathlib import Path

mistral_models_path = Path.home().joinpath('mistral_models', '7B-Instruct-v0.3')
mistral_models_path.mkdir(parents=True, exist_ok=True)

snapshot_download(repo_id="mistralai/Mistral-7B-Instruct-v0.3", allow_patterns=["params.json", "consolidated.safetensors", "tokenizer.model.v3"], local_dir=mistral_models_path)

To fetch the whole repository instead, use the standard Hugging Face commands.

pip install -U huggingface_hub
hf download mistralai/Mistral-7B-Instruct-v0.3

How to use it: After installing mistral_inference, the mistral-chat command opens a chat with the model from the folder you downloaded. If you prefer transformers, a text-generation pipeline takes a list of messages and returns the reply.

mistral-chat $HOME/mistral_models/7B-Instruct-v0.3 --instruct --max_tokens 256

from transformers import pipeline

messages = [
    {"role": "system", "content": "You are a pirate chatbot who always responds in pirate speak!"},
    {"role": "user", "content": "Who are you?"},
]
chatbot = pipeline("text-generation", model="mistralai/Mistral-7B-Instruct-v0.3")
chatbot(messages)

Tool | What it does
mistral-chat | Opens a command-line chat with the downloaded model
mistral_inference | The library the card recommends for running the model
transformers pipeline | A second way to generate text from a list of messages
snapshot_download | Fetches the three model files the card lists

Example: You describe a function named get_current_weather with a location and a temperature unit, then ask about the weather in Paris. The model is meant to answer with a call to that function, with the city filled in.

For newcomers and experts: A good first try is the mistral-chat command with a short question, such as asking for a brief explanation of machine learning. Function calling comes after that.

For experts, the card recommends mistral_inference but also gives transformers snippets. The examples decode with a temperature of 0.0 and a token limit of 64, and the download fetches params.json, consolidated.safetensors and a tokenizer.model.v3 file. A function is passed to the request as a tool, with a name, a description and a parameter schema. Through transformers, function calling needs version 4.42.0 or higher.

Download the three model files
Load the tokenizer and the model
Encode your chat request and any tools
Generate tokens and decode the reply

Good to know:

Title: Model facts
Item | Value
Parameters | 7.2B
Weights in BF16 | about 14.5 GB
Weights at 4-bit | about 3.6 GB, an estimate
Vocabulary | 32768
Format | Safetensors
Built on | Mistral-7B-v0.3

• Function calling: you describe a tool with a name, a description and its parameters, and the model can answer with a call to it.
• Install: the card recommends mistral_inference and also gives transformers snippets.
• Variants: the Hugging Face page lists many quantized and fine-tuned versions made by other people.
• License: apache-2.0
• Cost to run: no API key or paid service named. You need hardware for a 7.2B model.
• Data it reaches: it downloads files from Hugging Face, and the examples run the model on your own machine.
• Not for: none named.

On Hugging Face, 13 November 2026

• Library: vllm
• Parameters: 7.2B
• Weights: about 14.5 GB in BF16; about 3.6 GB at 4-bit (estimate)
• Formats: Safetensors
• Downloads in the last 30 days: 1,771,788
• Likes: 3,838
• License: apache-2.0
• Based on: mistralai/Mistral-7B-v0.3
• Quantized versions: 315
• Fine-tunes: 530
• Tags: vllm, safetensors, mistral, mistral-common

Links

• Model: https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.3
• Files: https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.3/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download mistralai/Mistral-7B-Instruct-v0.3

Then follow the usage in its model card. Nothing in this kit runs by itself.