Repository navigation
Revert "[Bugfix] Tie lm_head.weight for Nemotron Parse when checkpoint omits it" (#53020) - #59805
Merged
Merged
Conversation
…t omits it" (vllm-project#53020) This reverts commit 83cadd6. The new nvidia/NVIDIA-Nemotron-Parse-2.0 case in test_nemotron_parse.py fails on every run: its tokenizer fails to load with "DuplicatePatternError ... at line 449472 column 1" in TokenizerFast.from_file. Reverting until it can be fixed. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Qa3mjR8Niyg7WjvckpNRiu Signed-off-by: Thang Nguyen <thangnguyenvn647@gmail.com>
![claude[bot]](https://avatars.githubusercontent.com/in/1236702?s=60&v=4){kind=small}
mergify
Bot
added
multi-modality
DarkLight1337
approved these changes
Oct 2, 2026
6 tasks done
HangruiCao2000
pushed a commit
to HangruiCao2000/vllm
that referenced
this pull request
Oct 6, 2026
Re-lands vllm-project#53020 (commit 83cadd6), which was reverted in vllm-project#59805 because the NVIDIA-Nemotron-Parse-2.0 test case it added failed in CI while loading that checkpoint's tokenizer: transformers/tokenization_utils_tokenizers.py:120 in convert_to_native_format local_kwargs["tokenizer_object"] = TokenizerFast.from_file(fast_tokenizer_file) E Exception: DuplicatePatternError: [225, 131, 156] at line 449472 column 1 The revert asked for the fix to be re-landed "together with a working way to load that tokenizer". vllm-project#59796 provided it 74 minutes later by bumping tokenizers to >= 0.23.2, which main now pins, so the blocker is gone. The model change is byte-identical to the reverted commit: it restores the tie between lm_head and the decoder's input embeddings for checkpoints that set tie_word_embeddings and ship no lm_head.weight tensor. Without it the output head keeps its random initialization and the server returns garbage while reporting healthy. Signed-off-by: Anish Kumar <aniskumar@nvidia.com> Co-authored-by: Cyrus Leung <tlleungac@connect.ust.hk> Signed-off-by: DiegoCao <DiegoCao@users.noreply.github.com>
Merged
6 tasks done
aniskumar-nv
added a commit
to aniskumar-nv/vllm
that referenced
this pull request
Oct 7, 2026
…arse when checkpoint omits it Reverts vllm-project#59805 (7cfd233), restoring vllm-project#53020 (83cadd6) unchanged. nvidia/NVIDIA-Nemotron-Parse-2.0 sets tie_word_embeddings=True and ships no lm_head.weight; without the tie, NemotronParseForConditionalGeneration keeps a randomly initialized output head and every response is garbage while the server reports healthy. vllm-project#53020 was reverted because the Parse-2.0 test case it added failed in the nightly full CI run at tokenizer load (TokenizerFast.from_file -> DuplicatePatternError), before any model code ran. The cause was the CI pin tokenizers==0.23.1, which rejects added tokens that collide after NFKC normalization (14 such pairs in this tokenizer). vllm-project#59796 bumped the pin to 0.23.2, which loads the file, so the test case can run as written. Co-authored-by: Cyrus Leung <tlleungac@connect.ust.hk> Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Signed-off-by: Anish Kumar <aniskumar@nvidia.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Overview
Reverts #53020, at @DarkLight1337's request. The
nvidia/NVIDIA-Nemotron-Parse-2.0case it added totest_nemotron_parse.pyfails on every CI run, and the step has been red on main since it merged.Claims
Validation
test_models[5-bfloat16-nvidia/NVIDIA-Nemotron-Parse-2.0]The revert is the exact inverse of 83cadd6 (
git diff 83cadd65d9 83cadd65d9^matches this diff).ruff checkandruff format --checkpass on both files.Details
The failure happens while the 2.0 checkpoint's tokenizer is loading, before any model code runs:
So the new test case can't run with the tokenizer library in the CI image. Re-landing the
tie_weights()fix should go together with a working way to load that tokenizer, or without the 2.0 test case. cc @aniskumar-nvPull Request Checklist
I used vLLM's
/pr-checklistskill. (Mandatory for agents, optional for humans).AI assistance was used during the creation of this PR.
Design Fit: Minimizes impact on core components, reuses existing functionality, and justifies added complexity.
Testing and Validation: Validates the change and ensures any added tests are meaningful and reliable, with CI coverage or documented CI resource constraints and validation performed outside CI.
Code Quality and Style: Keeps code and comments clear and concise, and updates relevant documentation and examples.
Pull Request Contents: Includes a brief summary and relevant links, supports claims with evidence, explains root causes and implementation trade-offs, and follows the contributing guide.
🤖 Generated with Claude Code
https://claude.ai/code/session_01Qa3mjR8Niyg7WjvckpNRiu