Qwen3 4B / 8B / 14B at ~2.7 bits per weight with Leech lattice quantization. Beats IQ2_S on quality at the same size, loses on speed [open source, Rust + CUDA]

1. 

   消息推送

   Qwen3 4B / 8B / 14B at ~2.7 bits per weight with Leech lattice quantization. Beats IQ2_S on quality at the same size, loses on speed [open source, Rust + CUDA]

   51分钟前

   Qwen3 4B / 8B / 14B at ~2.7 bits per weight with Leech lattice quantization. Beats IQ2_S on quality at the same size, loses on speed [open source, Rust + CUDA]
   
   🌐 链接: https://www.reddit.com/r/LocalLLaMA/comments/1xwerbo/qwen3_4b_8b_14b_at_27_bits_per_weight_with_leech/
   🏷️ 分组: Reddit｜AI·自托管·VPS
   🕒 时间: 2026-11-15 06:10:59

   Reddit

   

   From the LocalLLaMA community on Reddit

   Explore this post and more from the LocalLLaMA community

   51分钟前