Mamba Codestral 7 B v0.1

Introduction

The Mamba-Codestral-7B-v0.1 is an open-source code model based on the Mamba2 architecture, designed to compete with state-of-the-art Transformer-based code models. It is developed by Mistral AI and performs efficiently on various industry-standard benchmarks.

Architecture

The model is built on the Mamba2 architecture, which is tailored for code generation and related tasks. This architecture allows the model to handle complex programming queries effectively, providing a robust solution for code-based applications.

Training

Mamba-Codestral-7B-v0.1 has been evaluated on several benchmarks, including HumanEval, MBPP, Spider, and others. The model demonstrates impressive performance across various datasets, indicating its capability to generate accurate and efficient code across different programming languages and scenarios.

Guide: Running Locally

1. Installation
   Begin by installing the necessary packages. You can use mistral_inference or the original mamba package:

   pip install mistral_inference>=1 mamba-ssm causal-conv1d
   

   or

   pip install mamba_ssm causal-conv1d
   

2. Download the Model
   Use the Hugging Face hub to download the model:

   from huggingface_hub import snapshot_download
   from pathlib import Path
   
   mistral_models_path = Path.home().joinpath('mistral_models', 'Mamba-Codestral-7B-v0.1')
   mistral_models_path.mkdir(parents=True, exist_ok=True)
   
   snapshot_download(repo_id="mistralai/Mamba-Codestral-7B-v0.1", allow_patterns=["params.json", "consolidated.safetensors", "tokenizer.model.v3"], local_dir=mistral_models_path)
   

3. Run the Model
   After installation, use the mistral-demo CLI to start a chat session:

   mistral-chat $HOME/mistral_models/Mamba-Codestral-7B-v0.1 --instruct --max_tokens 256
   

4. Cloud GPUs
   For optimal performance, consider using cloud-based GPU services such as AWS, Google Cloud, or Azure to handle the computational requirements of the model.

License

The Mamba-Codestral-7B-v0.1 is released under the Apache 2.0 license, which allows for wide use and modification while ensuring attribution to the original creators.