All pull requests

Pull requests

• GPU: add opt-in GFN-16/17 normalization workgroup tuning

  #30·cherubrock-seb opened Nov 8, 2026·

  ·

• GPU: add opt-in AMD GFN-21 8+8+5 plan

  #29·cherubrock-seb opened Nov 8, 2026·

  ·

• Fix TUNE forward4 multiplicand wrapper

  #28·cherubrock-seb opened Nov 8, 2026·

  ·

• GPU: add opt-in AMD GFN-22 8+8+6 plan

  #27·cherubrock-seb opened Nov 8, 2026·

  ·

• GPU: add opt-in GFN-17 and NVIDIA GFN-21 optimizations

  #26·cherubrock-seb opened Nov 7, 2026·

  ·

• Apple GPUs: the scale of normalize1 is folded into RSQ and MFIM

  #25·hairez opened Nov 4, 2026Contributor·

  ·

• Apple GPUs: blocks of 4 coefficients are squared directly

  #24·hairez opened Nov 4, 2026Contributor·

  ·

• Apple GPUs: normalize2 is merged into the next forward64_0

  #23·hairez opened Nov 4, 2026Contributor·

  ·

• Apple GPUs: Shoup multiplication by constants

  #22·hairez opened Nov 4, 2026Contributor·

  ·
  2