[Bugfix][Quantization][MoE] Route weight only NVFP4 checkpoints through W4A16 (#54427)

Signed-off-by: Aditya Jha <4adityajha@gmail.com>
Co-authored-by: Claude <noreply@anthropic.com>
This commit is contained in:
Aditya Jha
2026-08-30 13:02:15 -04:00
committed by GitHub
co-authored by Claude
parent e79961c946
commit 9d0fe9bac8
@@ -1081,6 +1081,22 @@ class ModelOptNvFp4Config(ModelOptQuantConfigBase):
) -> "ModelOptNvFp4Config":
is_checkpoint_nvfp4_serialized = "NVFP4" in quant_method
# A checkpoint can declare quant_algo "NVFP4" yet quantize weights
# only (input_activations is null in every config group). The W4A4
# MoE path then folds an uninitialized input_scale that was never
# loaded and can zero every expert, so route it through W4A16.
if quant_method == "NVFP4":
config_groups = original_config.get("config_groups")
if (
isinstance(config_groups, dict)
and config_groups
and all(
isinstance(group, dict) and group.get("input_activations") is None
for group in config_groups.values()
)
):
quant_method = "W4A16_NVFP4"
if group_size is None:
group_size = 16 # Default value