2026-10-27 04:55
AI123
Unsloth为MiniMax-M3.1-Flash大模型发布2-bit量化版，体积缩减45%至143.9GB
10月27日，Unsloth发布了MiniMax-M3.1-Flash的2-bit GGUF量化版本。该模型基于MiniMax的428B参数MoE架构，量化后体积从原版降至143.9GB，缩减约45%。在Unsloth的评估中，该版本在七个领域保持91.5%的top-1 token一致性。
这一版本旨在降低自托管门槛，支持通过Unsloth Desktop和llama.cpp进行本地部署，并提供OpenAI兼容API，方便现有应用和智能体接入。Unsloth目前在GitHub拥有超过7.8万星标，Hugging Face上模型下载量超过5亿次，MiniMax-M3.1-Flash借此进入其开源开发者社区。
来源
- 2026-10-27 04:25 | X：MiniMax M3.1 Flash meets the @UnslothAI community. 🐙🦥✨ Unsloth’s Dynamic GGUFs bring our 428B MoE model to a 143.9GB 2-bit footprint, with 91....阅读原文
MiniMax M3.1 Flash meets the @UnslothAI community. 🐙🦥✨ Unsloth’s Dynamic GGUFs bring our 428B MoE model to a 143.9GB 2-bit footprint, with 91.5% top-1 token agreement in their evaluation across seven domains. That means a lower memory barrier to self-hosting, deployment through Unsloth Desktop and...