Issues

• turbo3 CPU dequant (dequantize_row_turbo3_0) skips the inverse WHT, so the to_float round-trip is broken (cosine ~0). Patch + repro inside.

  scrya-com/rotorquant#16 · Filter by author hexbinoct opened on Jun 25, 2026

• About MTP

  scrya-com/rotorquant#13 · Filter by author b1231227 opened on May 7, 2026
  1

• Can you add support for Gemma 4 models?

  scrya-com/rotorquant#12 · Filter by author danganbenpa opened on May 4, 2026

• [Bug] iso3/planar3 KV prefill scales super-linearly and produces garbage on RTX 5090 (sm_120 / CUDA 12.8)

  scrya-com/rotorquant#10 · Filter by author TanveerKahlon opened on Apr 21, 2026

• Unknown model architecture Gemma4 - Pull Gemma4 support from latest llama.cpp code

  scrya-com/rotorquant#9 · Filter by author timematcher opened on Apr 13, 2026
  1

• RotorQuant / PlanarQuant / IsoQuant extremely slow on Metal with high graph splits (>2)

  scrya-com/rotorquant#7 · Filter by author lvirbalas opened on Apr 2, 2026
  2

• Using rotorquant for model weights

  scrya-com/rotorquant#5 · Filter by author cnmoro opened on Mar 29, 2026

• AMD ROCm support

  scrya-com/rotorquant#1 · Filter by author hhenkel opened on Mar 26, 2026
  3