biondizzle 201a40e6c4 Fix zero-dim tensor concatenation in compressor scale buffer ...

input_scale and weight_scale_2 are 0-dim scalars in the NVFP4 checkpoint.
torch.cat can't concatenate scalars — reshape to 1-d first.

2026-05-19 00:10:13 +00:00

..

patches

Fix zero-dim tensor concatenation in compressor scale buffer

2026-05-19 00:10:13 +00:00

cutedsl_quant_method.py

Fix: add abstract create_weights to CuTeDSLNvfp4LinearMethod

2026-05-18 20:40:48 +00:00

nvfp4_cutedsl.py

Fix torch.compile: use custom autograd Function instead of @torch.compiler.disable

2026-05-18 21:38:28 +00:00