Mistral 7B
#47 en Modèles de langage open sourceMistral AI · 7b · depuis 27. September 2023 · 4× · vu le 09 août 2026
25
Momentum
Mistral 7B is an open-source language model with 7 billion parameters, released by Mistral AI in September 2023. It uses Grouped-Query Attention and Sliding Window Attention to achieve performance surpassing larger models like Llama 2 13B while requiring less compute. The model is text-only, supports multiple languages, and is available under the permissive Apache 2.0 license for free download and commercial use. It is frequently used as a base for fine-tuning projects, such as document summarization.
Historique du momentum
16.05.14.08.
Fonctionnalités
| Key Benchmark (%) | MMLU: 60.1% (5-shot), surpasses Llama 2 13B (55.6%) |
| Context Window (Tokens) | 8,192 tokens (sliding window attention with 4,096-token window) |
| License | Apache 2.0 (unrestricted use) |
| Multimodality | No, text-only model (no image/audio support) |
| Platform | Download/local, HuggingFace, Cloud (AWS/GCP/Azure) via vLLM/SkyPilot |
| Price per 1M Tokens | approx. $0.20 input / $0.20 output (third-party API pricing, varies by provider) |
| Release Date | September 27, 2023 |