biondizzle 103fd451ce fix: use full padded_scales_buf (no GPU scalar slicing in cudagraph) ...

buf[:gpu_scalar, :] triggers cudaErrorStreamCaptureInvalidated.
Always use the full pre-allocated buffer; extra rows are zeros.

2026-05-16 18:50:35 +00:00

..

patches

fix: remove broken hc_head warmup (wrong tensor shape)

2026-05-16 10:11:34 +00:00

nvfp4_cutedsl.py

fix: use full padded_scales_buf (no GPU scalar slicing in cudagraph)

2026-05-16 18:50:35 +00:00