biondizzle 94dec5922d Scale assembly Phase 2: use CPU-computed offsets for Python slicing ...

GPU scalars can't be used for Python indexing (requires sync).
Compute padded_expert_offsets on CPU via .cpu().tolist() for
the Python loop. This is OK for cudagraph: Python code only
runs during capture, not replay. The GPU kernel launches
recorded during capture are deterministic.

2026-05-17 16:56:52 +00:00

..

patches

Fix shape mismatch: shared padded buffers, revert max_num_tokens cap

2026-05-17 15:47:10 +00:00

nvfp4_cutedsl.py

Scale assembly Phase 2: use CPU-computed offsets for Python slicing

2026-05-17 16:56:52 +00:00