KQuant - new quantization technique, 3.7× better quants for Qwen Flash Next 180B, Qwen 27B, 35B, 9B (measured in KL-divergence reduction) (MLX)

🌐 链接: https://www.reddit.com/r/LocalLLaMA/comments/1xu9onk/kquant_new_quantization_technique_37_better/
🏷️ 分组: Reddit｜AI·自托管·VPS
🕒 时间: 2026-11-12 18:47:24
Reddit
From the LocalLLaMA community on Reddit: KQuant - new quantization technique, 3.7× better quants for Qwen Flash Next 180B, Qwen…
From the LocalLLaMA community on Reddit: KQuant - new quantization technique, 3.7× better quants for Qwen Flash Next 180B, Qwen…
Explore this post and more from the LocalLLaMA community