Pull requests / #706

#706 Feat/q2 avx2 bitplane

closed · @merbanan · 0 commentaires · Sur GitHub

Models & quants

Description

A new AVX2 row kernel for Q2_0 experts on CPUs without AVX-512, the default path in `q2_0_gguf_rows_multi_avx2`.

**Single thread, weights in L2:** 2.3–2.8× faster at every token count from 1 to 4.

Sur le site

Liens install, modèles, releases.