腾讯混元将 Hy4-preview 压缩至约 200GiB GGUF，发布 MIX-STQ1_0 量化方案
We compressed Hy4-preview from 1.5TB to ～200GiB GGUF and it still works well !
腾讯混元把 Hy4-preview 从 1.5TB 压缩到约 200GiB 的 GGUF，并称仍能良好运行。其 MIX-STQ1_0 方案用校准数据为每层选择位宽，部分低至 1.31-bit STQ1_0，部分高至 2.06-bit IQ2_XXS。相比 BF16，MCP Atlas 从 83.7 降至 83.2，SWE-Bench multi 从 82.9 降至 81.3，MRCR 从 81.3 降至 81.1，IFBench 从 73.5 降至 72.5。权重与低比特 GGUF 已发布在 Hugging Face 的 AngelSlim/Hy4-preview-GGUF。
推荐理由
AI 评分 62
腾讯混元给出 Hy4-preview 低比特量化的分层位宽方案与四项基准对比，可据此判断本地部署的精度代价。
主题
查看原帖🚀 Hy4 preview is here. 770B, 49B active, 1M context. Built for productivity. Open source frontier. Consistent affordable price. Use it. Tell us what breaks. More on Hy blog：https://t.co/rbl1IWRk3C HuggingFace：https://t.co/mE9wevH5XR Github：https://t.co/pyl9zckpoL https://t.co/4iW6gSuZKr
正文
当前显示原文，中文全文译文尚未就绪。
We compressed Hy4-preview from 1.5TB to ～200GiB GGUF and it still works well !
Meet MIX-STQ1_0.The trick isn’t just going low, it’s deciding where: calibration data picks each layer’s bit-width, some down to 1.31-bit STQ1_0, some up to 2.06-bit IQ2_XXS. Same budget, lower error.
Accuracy barely moves vs BF16 📊 MCP Atlas 83.7→83.2 📊 SWE-Bench multi 82.9→81.3 📊 MRCR 81.3→81.1 📊 IFBench 73.5→72.5
See the details on HF : AngelSlim/Hy4-preview-GGUF
Weights & low-bit GGUFs 👇 https://t.co/9uM9NT9Wem
#LLM #Quantization #llamacpp #Hy