Add bagel step in the test-nightly.yml#1
Merged
natureofnature merged 1 commit intoApr 17, 2026
Merged
Conversation
Signed-off-by: NumberWan <wantszkin2003@gmail.com>
natureofnature
added a commit
that referenced
this pull request
May 21, 2026
Address issue on 809f6e1 (v5): - High #1: build_pooler_payload received `out_idx` but sampler_output/invalid_req_indices index by input_batch row. - High vllm-project#2: gpu_generation_model_runner._overlay_full_payload_input_ids was CosyVoice3-specific in the common runner. - Medium vllm-project#3: cosyvoice3._pooler_output_history_from_input_batch didn't stop at -1 placeholder. Resolution: drop the connector codec path for CosyVoice3 sync and deliver codec via legacy `additional_information`, strip prompt/reference prefix at the SIP layer, and gate code2wav mel-trim on the talker-prefill offset only when a speech-stop token was seen. Source changes: - gpu_ar_model_runner.py: remove build_pooler_payload hook + _attach_model_pooler_payload + _pooler_payload_has_key. - gpu_generation_model_runner.py: remove _flatten_audio_codes_to_tensor + _overlay_full_payload_input_ids + its call site. - cosyvoice3.py (model): remove build_pooler_payload + _pooler_codec_rows + _pooler_output_history_from_input_batch + _pooler_sampled_token_ids (and three per-req caches: _pooler_codec_history_by_req, _pooler_codec_sampled_seen_by_req, _pooler_codec_sampled_finished_by_req). code2wav.forward token_offset_tokens now reads `meta.talker_prefill_offset` (already a struct field used by qwen3_tts). - SIP cosyvoice3.py: text2flow + text2flow_token_only strip the prompt token prefix and the prompt speech_token prefix from cumulative_token_ids; set meta.talker_prefill_offset only when raw output contains a speech-stop token. text2flow_full_payload no longer ships codes.audio (embed/meta only). Drop `codes.audio` from _FULL_PAYLOAD_REPLACE_KEYS. - _to_token_id_list no longer filters negative ids (needed for stop-token detection on raw cumulative ids). Side effects: - v5's cosyvoice3 per-req cache leak is gone (no pooler hook → no accumulator). - The pre-existing baseline `voice_clone_zh_001[cosyvoice3]` sim=0.00 (transcript "先") failure is fixed. Verification on H800 GPU with `--run-level full_model -m "full_model and tts"`: - test_voice_clone_zh_001[cosyvoice3]: PASS sim=1.000 (baseline FAIL sim=0.00; v5 PASS sim=0.903) - test_voice_clone_en_001[cosyvoice3]: PASS sim=0.963 (baseline PASS sim=0.946; v5 PASS sim=0.963) Trade-off vs project_pr3_scope: CosyVoice3 sync codec stays on legacy additional_information; embed/prompt conditioning still ships via connector. Other PR3-migrated archs are unaffected (none consumed codes.audio via the removed overlay). Signed-off-by: natureofnature <wzliu@connect.hku.hk>
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.
PLEASE FILL IN THE PR DESCRIPTION HERE ENSURING ALL CHECKLIST ITEMS (AT THE BOTTOM) HAVE BEEN CONSIDERED.
Purpose
Test Plan
Test Result
Essential Elements of an Effective PR Description Checklist
supported_models.mdandexamplesfor a new model. Please runmkdocs serveto sync the documentation editions to./docs.BEFORE SUBMITTING, PLEASE READ https://github.com/vllm-project/vllm-omni/blob/main/CONTRIBUTING.md (anything written below this line will be removed by GitHub Actions)