biondizzle cdd813cf7e fix: free per-expert weight lists after stacking in CuTeDSL runner ...

_ensure_stacked() creates stacked copies of all weights but never freed
the per-expert lists. For 256 experts on a 175GB model, this doubles
weight memory to ~350GB, causing OOM.

Now the per-expert lists (l1_fp4, l1_sf, l1_gs, l2_fp4, l2_sf, l2_gs)
are set to None after stacking, keeping only the single stacked copy.

2026-05-16 08:54:52 +00:00

..

patches

Add tilelang kernel warmup in load_weights

2026-05-16 08:28:39 +00:00

nvfp4_cutedsl.py

fix: free per-expert weight lists after stacking in CuTeDSL runner

2026-05-16 08:54:52 +00:00