Mistral - AshokBhat/ml GitHub Wiki

About

  • Mistral AI, a French company
  • Founded in April 2023, by former employees of Meta Platforms and Google DeepMind
  • Has released open source models (Apache 2.0 license)

Models

Model Size Release date Details
Mistral-7B 7B Sep 23 First model family from mistral.ai
Mixtral-8x7B 12B active Dec 23 Sparse Mixture-of-Experts model

Variants

Model Name Description
<>-v0.1 Base model
<>-Instruct-v0.1 Instruction-tuned model optimized for chat purposes
<>-Instruct-v0.2 Improved instruction-tuned model, building upon v0.1.

Mixtral-8x7B Performance

  • Outperforms Llama2 70B on most benchmarks with 6x faster inference.
  • Matches or outperforms GPT3.5 on most standard benchmarks.

See also