biondizzle 5c770c68ca Keep MoE scale tensors: framework warmup needs them ...

The framework's deep_gemm_warmup calls get_fused_moe_quant_config
which accesses w13_input_scale etc. Setting them to None caused
TypeError: float / NoneType. Keep scales (small tensors) and only
free the large weight tensors.

2026-05-19 04:50:31 +00:00

..

kernels/linear/nvfp4

Replace autograd.Function with torch.library.custom_op for Dynamo compat

2026-05-19 01:54:48 +00:00

patches

Keep MoE scale tensors: framework warmup needs them

2026-05-19 04:50:31 +00:00

cutedsl_quant_method.py

Replace autograd.Function with torch.library.custom_op for Dynamo compat

2026-05-19 01:54:48 +00:00

nvfp4_cutedsl.py

Replace autograd.Function with torch.library.custom_op for Dynamo compat

2026-05-19 01:54:48 +00:00