Codestral Mamba is the latest addition to our architecture research efforts, providing a free, customizable model designed by Albert Gu and Tri Dao. Unlike Transformer models, Mamba models offer linear time inference and can theoretically model sequences of infinite length, making it highly efficient for code productivity tasks. Tested on in-context retrieval capabilities up to 256k tokens, Codestral Mamba is expected to be a powerful local code assistant. You can deploy it using various tools like mistral-inference SDK or TensorRT-LLM. Available under the Apache 2.0 license, it boasts an impressive 7,285,403,648 parameters. Don’t forget to check out Codestral 22B, its commercial or community licensing options!
https://mistral.ai/news/codestral-mamba/