unsloth/Magistral-Small-2507
Magistral-Small-2507 is a 24 billion parameter language model developed by Mistral AI, built upon the Mistral Small 3.1 architecture. It is specifically enhanced for reasoning capabilities, incorporating SFT from Magistral Medium traces and RL. This multilingual model excels at complex reasoning tasks and supports a 40k token context window, making it suitable for applications requiring deep analytical processing.
Loading preview...
Magistral Small 1.1 Overview
Magistral Small 1.1 is a 24 billion parameter model from Mistral AI, designed for efficient reasoning. It builds upon the Mistral Small 3.1 architecture, incorporating supervised fine-tuning (SFT) from Magistral Medium traces and reinforcement learning (RL) to enhance its analytical capabilities. This version improves upon Magistral Small 1.0 with better tone, LaTeX and Markdown formatting, and reduced likelihood of infinite generation loops.
Key Capabilities
- Advanced Reasoning: Capable of generating long chains of reasoning traces before formulating an answer, aided by
[THINK]and[/THINK]special tokens for structured thought processes. - Multilingual Support: Proficient in dozens of languages, including English, French, German, Japanese, Chinese, and Arabic.
- Optimized Context Window: Supports a 128k context window, with recommended optimal performance up to 40k tokens.
- Deployment Flexibility: Designed to be deployable locally, fitting on hardware like a single RTX 4090 or a 32GB RAM MacBook when quantized.
- Apache 2.0 License: Allows for broad commercial and non-commercial use and modification.
Good For
- Applications requiring robust reasoning and problem-solving.
- Multilingual text generation and understanding.
- Tasks benefiting from structured thought processes and detailed explanations.
- Local deployments on consumer-grade hardware for efficient inference.