This is a 4-bit AWQ quantized version of a 24B parameter instruct model derived from the Mistral3 architecture. It is optimized for local inference and excels at coding and technical tasks. The model is distributed on Hugging Face for use with Transformers or compatible inference engines.
In the Quantised & converted weights space, Devstral Small 2 24B Instruct 2512 takes a focused approach. It focuses on running a capable coding-focused large language model locally with reduced memory requirements. It is built as an open-source project for developers. The project is open source (Apache-2.0). Devstral Small 2 24B Instruct 2512 is available on the web, the command line, and API.
cyankiwi builds and maintains Devstral Small 2 24B Instruct 2512, and it first shipped in 2025. The project is developed in the open on GitHub with 4.7k stars and 32 commits in the last 90 days. Key capabilities include instruct-tuned, 4-bit Quantized, and Code Generation.
Summary written by a language model from the project’s public pages.
What PulseGate has recorded for this listing
Closest matches by what these projects do