Skip to content

Revert "[Bugfix] Tie lm_head.weight for Nemotron Parse when checkpoint omits it" (#53020) #59805

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
vllm-bot merged 1 commit into from
Oct 2, 2026
Merged

Revert "[Bugfix] Tie lm_head.weight for Nemotron Parse when checkpoint omits it" (#53020) #59805

Show file tree
Hide file tree
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
11 changes: 1 addition & 10 deletions tests/models/multimodal/generation/test_nemotron_parse.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -104,16 +104,7 @@ def run_test(
)


@pytest.mark.parametrize(
"model",
[
"nvidia/NVIDIA-Nemotron-Parse-v1.2",
# v2.0's checkpoint ties lm_head to the decoder's input embeddings
# instead of shipping a separate lm_head.weight tensor; this
# exercises that path (see tie_weights() in nemotron_parse.py).
"nvidia/NVIDIA-Nemotron-Parse-2.0",
],
)
@pytest.mark.parametrize("model", ["nvidia/NVIDIA-Nemotron-Parse-v1.2"])
@pytest.mark.parametrize("dtype", ["bfloat16"])
@pytest.mark.parametrize("num_logprobs", [5])
def test_models(
Expand Down
9 changes: 0 additions & 9 deletions vllm/model_executor/models/nemotron_parse.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -568,15 +568,6 @@ def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
self.lm_head = ParallelLMHead(
config.decoder.vocab_size, config.decoder.d_model, quant_config=quant_config
)
# Some checkpoints (e.g. compact exports) tie the output head to the
# decoder's input embeddings instead of materializing a separate
# lm_head.weight tensor.
tie_word_embeddings = bool(
getattr(config, "tie_word_embeddings", False)
or getattr(config.decoder, "tie_word_embeddings", False)
)
if tie_word_embeddings:
self.lm_head = self.lm_head.tie_weights(self.decoder.embed_tokens)
self.logits_processor = LogitsProcessor(
self.vocab_size, config.decoder.vocab_size
)
Expand Down
Loading