perf(gemma4 MTP H100): tune Triton extend tile for Lq=256 / sm_90#4
Draft
pyc96 wants to merge 2 commits into
Draft
perf(gemma4 MTP H100): tune Triton extend tile for Lq=256 / sm_90#4pyc96 wants to merge 2 commits into
pyc96 wants to merge 2 commits into