Repository navigation
[LoRA] Cleanup VocabParallelEmbedding - #53843
Merged
Merged
Conversation
![claude[bot]](https://avatars.githubusercontent.com/in/1236702?s=60&v=4){kind=small}
There was a problem hiding this comment.
Claude Code Review
This repository is configured for manual code reviews. Comment @claude review for a one-time review, or @claude review always to subscribe this PR to a review on every future push.
Tip: disable this comment in your organization's Code Review settings.
Member
Author
|
/ci run |
|
✅ Triggered Buildkite CI #85599 for commit |
Isotr0py
approved these changes
Aug 26, 2026
pawel-olejniczak
added a commit
to pawel-olejniczak/vllm-gaudi
that referenced
this pull request
Aug 27, 2026
…flattened LoRA _embeddings_indices Root cause: vLLM PR #53843 flattened PunicaWrapperBase._embeddings_indices from [2, max_num_batched_tokens] to 1-D, so the HPU override's [1] select yielded a 0-dim tensor and slicing it raised IndexError. Upstream: vllm-project/vllm#53843 (75aa189) Fix: drop the stale row select and slice the 1-D indices directly, mirroring upstream's own forward update. Signed-off-by: Paweł Olejniczak <pawelx.olejniczak@intel.com>
iboiko-habana
pushed a commit
to vllm-project/vllm-gaudi
that referenced
this pull request
Aug 28, 2026
…PU HunYuan, Import cli_args from (+3 more) (#1759) This PR consolidates 5 hourly-CI fixes against vllm@`17da48596c98946d3e3e6896e2ebd341e809f3bd` per the single-rolling-PR rule (invariant I9). ## Bug 1: Drop HPU HunYuan V1 overrides deleted upstream - **State machine id**: hunyuan_v1_model_module_removed_plugin_reg - **Commit**: c352fb7 ### Root cause vllm#53615 (48d7132) migrated HunYuan V1/VL to the Transformers modeling backend and deleted vllm/model_executor/models/hunyuan_v1.py, so the eager import in vllm_gaudi/models/__init__.py aborted plugin registration and killed 48 of 51 jobs in CI run 32990520142 before any test could run. ### Culprit Regression introduced by [PR #53615](vllm-project/vllm#53615). ### Fix remove vllm_gaudi/models/hunyuan_v1.py and its two register_model calls; upstream's registry now serves both architectures via the Transformers backend and the HPU subclass only specialized the now-deleted HunYuanAttention. ## Bug 2: Import cli_args from vllm.entrypoints.launchers - **State machine id**: entrypoints_openai_cli_args_module_moved - **Commit**: 935d821 ### Root cause upstream moved vllm/entrypoints/openai/cli_args.py to vllm/entrypoints/launchers/cli_args.py with no compatibility re-export, so multi_model_api_server.py raised ModuleNotFoundError at import time and took down run_unit_tests plus e2e run_online_model_swap_test. ### Culprit Regression introduced by [PR #53659](vllm-project/vllm#53659). ### Fix import make_arg_parser and validate_parsed_serve_args from the new vllm.entrypoints.launchers.cli_args path; both symbols are unchanged there. ## Bug 3: Rename group_mm_kwargs_by_modality to group_and_batch_mm_kwargs - **State machine id**: group_mm_kwargs_by_modality_import_removed - **Commit**: 0988519 ### Root cause vllm#53559 (4f686e1) deleted the deprecated group_mm_kwargs_by_modality alias, so importing HPUModelRunner raises ImportError. ### Culprit Regression introduced by [PR #53559](vllm-project/vllm#53559). ### Fix import and call group_and_batch_mm_kwargs, the signature- and yield-identical replacement upstream renamed it to in vllm#36158. ## Bug 4: Index flattened LoRA _embeddings_indices - **State machine id**: hpu_lora_embeddings_indices_zero_dim_indexerror - **Commit**: fb5e4b0 ### Root cause vLLM PR #53843 flattened PunicaWrapperBase._embeddings_indices from [2, max_num_batched_tokens] to 1-D, so the HPU override's [1] select yielded a 0-dim tensor and slicing it raised IndexError. ### Culprit Regression introduced by [PR #53843](vllm-project/vllm#53843). ### Fix drop the stale row select and slice the 1-D indices directly, mirroring upstream's own forward update. ## Bug 5: Override synchronize_device for HPU - **State machine id**: synchronize_device_torch_accelerator_unsupported - **Commit**: e01570f ### Root cause vllm#52914 added WorkerBase.synchronize_device() defaulting to torch.accelerator.synchronize() plus an EngineCore._finish_pause() that collective_rpc's it on every pause, and HPU rejects a device-wide wait. ### Culprit Regression introduced by [PR #52914](vllm-project/vllm#52914). ### Fix HPUWorker.synchronize_device() waits via torch.hpu.synchronize(). --------- Signed-off-by: Paweł Olejniczak <pawelx.olejniczak@intel.com>
am-cohere
pushed a commit
to am-cohere/vllm
that referenced
this pull request
Sep 1, 2026
Signed-off-by: Jee Jee Li <jeejeelee@inferact.ai>
mikeshawcode
pushed a commit
to mikeshawcode/vllm
that referenced
this pull request
Sep 1, 2026
Signed-off-by: Jee Jee Li <jeejeelee@inferact.ai> Signed-off-by: mikeshawcode <michaelwshaw2@gmail.com>
mylibrar
pushed a commit
to tanyuqian/vllm
that referenced
this pull request
Sep 3, 2026
Signed-off-by: Jee Jee Li <jeejeelee@inferact.ai>
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
2 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.
Purpose
We removed the feature of extending vocab size with LoRA a long time ago; this PR cleans up the related unused code.
Test Plan
Test Result
Essential Elements of an Effective PR Description Checklist
supported_models.mdandexamplesfor a new model.