Codestral Mamba 7B v0.1
Mistral AI · 2024-07-16 · 7.3B parameters
Codestral Mamba is Mistral AI's 7.3B-parameter instructed code model built on the Mamba-2 state-space architecture rather than a Transformer, released on July 16, 2024 with Apache 2.0 weights. Mistral designed it with help from Albert Gu and Tri Dao for linear-time inference, tested in-context retrieval up to 256k tokens, and served it on la Plateforme as codestral-mamba-2407.
Benchmark scores
| Benchmark | Score |
|---|---|
| HumanEval | 75.0 |
| MBPP | 68.5 |