2026-10-18 · America/Los_Angeles · 社区动态 · #7

New quant types suggested: IQ2_K, IQ3_K, IQ4_K, IQ5_K and IQ6_K

Looks quite interesting. Lots of numbers presented, but for a single model only. Highlights: File sizes are the same as Q-quants. KLD improves substantially over Q-quants. Performance suffers slightly Creating IQ quants is somewhat slow. Asterisks apply... https://github.com/ggml-org/llama.cpp/pull/30715 https://huggingface.co/eaddario/IQ2_K-IQ6_K-Test-Models Appologies to Mr. Addario if I mangled anything.

热度 47.5 / 100；排名与评分保留该期记录。