Skip to content

[TRTLLM-11768][fix] Config updates to enable NVFP4 #12776

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
2ez4bz merged 1 commit into from
Apr 7, 2026
Merged

[TRTLLM-11768][fix] Config updates to enable NVFP4 #12776

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
27 changes: 27 additions & 0 deletions tensorrt_llm/_torch/models/modeling_nemotron_nano.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1251,6 +1251,8 @@ def __init__(self, model_config: ModelConfig):

llm_model_config = copy.deepcopy(model_config)
llm_model_config.pretrained_config = llm_model_config.pretrained_config.llm_config
self._update_config_for_quantization(llm_model_config)

self.llm = AutoModelForCausalLM.from_config(llm_model_config)

self.vocab_size = llm_model_config.pretrained_config.vocab_size
Expand Down Expand Up @@ -1467,6 +1469,31 @@ def forward(
logger.debug(f"output shape: {output_prob.shape}")
return output_prob

@staticmethod
def _update_config_for_quantization(llm_model_config: ModelConfig) -> None:
# Strip the VL wrapper prefix from exclude_modules and
# quant_config_dict so patterns match the inner LLM's module names
# (e.g. "language_model.backbone.layers.0.mixer.conv1d" becomes
# "backbone.layers.0.mixer.conv1d").
_LM_PREFIX = "language_model."
if llm_model_config.quant_config.exclude_modules is not None:
llm_model_config.quant_config.exclude_modules = [
m[len(_LM_PREFIX) :] if m.startswith(_LM_PREFIX) else m
for m in llm_model_config.quant_config.exclude_modules
]
if llm_model_config.quant_config_dict is not None:
# NOTE: without `_frozen` toggling, `ModelConfig` cannot have its attributes
# modified.
old_frozen = llm_model_config._frozen
llm_model_config._frozen = False
try:
llm_model_config.quant_config_dict = {
k[len(_LM_PREFIX) :] if k.startswith(_LM_PREFIX) else k: v
for k, v in llm_model_config.quant_config_dict.items()
}
finally:
llm_model_config._frozen = old_frozen


def _rearrange_img(x: torch.Tensor, patch_size: int) -> torch.Tensor:
py = x.shape[-2] // patch_size
Expand Down
Loading