Skip to content

[Bugfix][Qwen3-TTS] Add compatibility for both @check_model_inputs and @check_model_inputs().#965

Closed
gcanlin wants to merge 3 commits intovllm-project:mainfrom
gcanlin:check_inputs
Closed

[Bugfix][Qwen3-TTS] Add compatibility for both @check_model_inputs and @check_model_inputs().#965
gcanlin wants to merge 3 commits intovllm-project:mainfrom
gcanlin:check_inputs

Conversation

@gcanlin
Copy link
Contributor

@gcanlin gcanlin commented Jan 26, 2026

PLEASE FILL IN THE PR DESCRIPTION HERE ENSURING ALL CHECKLIST ITEMS (AT THE BOTTOM) HAVE BEEN CONSIDERED.

Purpose

Fix #963.

Test Plan

python end2end.py --query-type Base
python end2end.py --query-type Base --use-batch-sample

Test Result

INFO 01-26 09:13:01 [log_utils.py:550] {'type': 'request_level_metrics',
INFO 01-26 09:13:01 [log_utils.py:550]  'request_id': '0_a52daf69-e539-4590-bc54-372bccd25e40',
INFO 01-26 09:13:01 [log_utils.py:550]  'e2e_time_ms': 8056.328058242798,
INFO 01-26 09:13:01 [log_utils.py:550]  'e2e_tpt': 287.72600208009993,
INFO 01-26 09:13:01 [log_utils.py:550]  'e2e_total_tokens': 28,
INFO 01-26 09:13:01 [log_utils.py:550]  'transfers_total_time_ms': 0.0,
INFO 01-26 09:13:01 [log_utils.py:550]  'transfers_total_bytes': 0,
INFO 01-26 09:13:01 [log_utils.py:550]  'stages': {0: {'stage_gen_time_ms': 7991.328954696655,
INFO 01-26 09:13:01 [log_utils.py:550]                 'num_tokens_out': 0,
INFO 01-26 09:13:01 [log_utils.py:550]                 'num_tokens_in': 28}}}
Processed prompts: 100%|███████████████████████████████████| 1/1 [00:08<00:00,  8.06s/req, est. speed stage-0 tok/s: 3.48, avg e2e_lat: 0.0ms]
INFO 01-26 09:13:01 [omni.py:840] [Summary] {'e2e_requests': 1,1 [00:08<00:00,  8.06s/req, est. speed stage-0 tok/s: 3.48, avg e2e_lat: 0.0ms]
INFO 01-26 09:13:01 [omni.py:840]  'e2e_total_time_ms': 8058.4917068481445,
INFO 01-26 09:13:01 [omni.py:840]  'e2e_sum_time_ms': 8056.328058242798,
INFO 01-26 09:13:01 [omni.py:840]  'e2e_total_tokens': 28,
INFO 01-26 09:13:01 [omni.py:840]  'e2e_avg_time_per_request_ms': 8056.328058242798,
INFO 01-26 09:13:01 [omni.py:840]  'e2e_avg_tokens_per_s': 3.475528776581028,
INFO 01-26 09:13:01 [omni.py:840]  'wall_time_ms': 8058.4917068481445,
INFO 01-26 09:13:01 [omni.py:840]  'final_stage_id': {'0_a52daf69-e539-4590-bc54-372bccd25e40': 0},
INFO 01-26 09:13:01 [omni.py:840]  'stages': [{'stage_id': 0,
INFO 01-26 09:13:01 [omni.py:840]              'requests': 1,
INFO 01-26 09:13:01 [omni.py:840]              'tokens': 28,
INFO 01-26 09:13:01 [omni.py:840]              'total_time_ms': 8056.776285171509,
INFO 01-26 09:13:01 [omni.py:840]              'avg_time_per_request_ms': 8056.776285171509,
INFO 01-26 09:13:01 [omni.py:840]              'avg_tokens_per_s': 3.4753354206364127}],
INFO 01-26 09:13:01 [omni.py:840]  'transfers': []}
Adding requests:   0%|                                                                                                  | 0/1 [00:08<?, ?it/s]
Request ID: 0_a52daf69-e539-4590-bc54-372bccd25e40, Saved audio to output_audio/output_0_a52daf69-e539-4590-bc54-372bccd25e40.wav
INFO 01-26 09:15:53 [log_utils.py:550] {'type': 'request_level_metrics',
INFO 01-26 09:15:53 [log_utils.py:550]  'request_id': '1_d4515d52-59f1-4cc9-ad81-feabd1827eff',
INFO 01-26 09:15:53 [log_utils.py:550]  'e2e_time_ms': 17422.70851135254,
INFO 01-26 09:15:53 [log_utils.py:550]  'e2e_tpt': 791.94129597057,
INFO 01-26 09:15:53 [log_utils.py:550]  'e2e_total_tokens': 22,
INFO 01-26 09:15:53 [log_utils.py:550]  'transfers_total_time_ms': 0.0,
INFO 01-26 09:15:53 [log_utils.py:550]  'transfers_total_bytes': 0,
INFO 01-26 09:15:53 [log_utils.py:550]  'stages': {0: {'stage_gen_time_ms': 8563.502788543701,
INFO 01-26 09:15:53 [log_utils.py:550]                 'num_tokens_out': 0,
INFO 01-26 09:15:53 [log_utils.py:550]                 'num_tokens_in': 22}}}
Processed prompts: 100%|████████████████████████████████| 2/2 [00:17<00:00,  8.71s/req, est. speed stage-0 tok/s: 2.87, avg e2e_lat: 8863.3ms]
INFO 01-26 09:15:53 [omni.py:840] [Summary] {'e2e_requests': 2,00:17<00:00,  8.68s/req, est. speed stage-0 tok/s: 2.87, avg e2e_lat: 8863.3ms]
INFO 01-26 09:15:53 [omni.py:840]  'e2e_total_time_ms': 17424.197912216187,
INFO 01-26 09:15:53 [omni.py:840]  'e2e_sum_time_ms': 26285.984754562378,
INFO 01-26 09:15:53 [omni.py:840]  'e2e_total_tokens': 50,
INFO 01-26 09:15:53 [omni.py:840]  'e2e_avg_time_per_request_ms': 13142.992377281189,
INFO 01-26 09:15:53 [omni.py:840]  'e2e_avg_tokens_per_s': 1.9021543406823156,
INFO 01-26 09:15:53 [omni.py:840]  'wall_time_ms': 17424.197912216187,
INFO 01-26 09:15:53 [omni.py:840]  'final_stage_id': {'0_b9b00e74-355c-4b24-b707-d3dc55b11612': 0,
INFO 01-26 09:15:53 [omni.py:840]                     '1_d4515d52-59f1-4cc9-ad81-feabd1827eff': 0},
INFO 01-26 09:15:53 [omni.py:840]  'stages': [{'stage_id': 0,
INFO 01-26 09:15:53 [omni.py:840]              'requests': 2,
INFO 01-26 09:15:53 [omni.py:840]              'tokens': 50,
INFO 01-26 09:15:53 [omni.py:840]              'total_time_ms': 17423.3341217041,
INFO 01-26 09:15:53 [omni.py:840]              'avg_time_per_request_ms': 8711.66706085205,
INFO 01-26 09:15:53 [omni.py:840]              'avg_tokens_per_s': 2.8697148118003097}],
INFO 01-26 09:15:53 [omni.py:840]  'transfers': []}
Adding requests:   0%|                                                                                                  | 0/2 [00:17<?, ?it/s]
Request ID: 0_b9b00e74-355c-4b24-b707-d3dc55b11612, Saved audio to output_audio/output_0_b9b00e74-355c-4b24-b707-d3dc55b11612.wav
Request ID: 1_d4515d52-59f1-4cc9-ad81-feabd1827eff, Saved audio to output_audio/output_1_d4515d52-59f1-4cc9-ad81-feabd1827eff.wav

Essential Elements of an Effective PR Description Checklist
  • The purpose of the PR, such as "Fix some issue (link existing issues this PR will resolve)".
  • The test plan, such as providing test command.
  • The test results, such as pasting the results comparison before and after, or e2e results
  • (Optional) The necessary documentation update, such as updating supported_models.md and examples for a new model.
  • (Optional) Release notes update. If your change is user facing, please update the release notes draft.

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)

@gcanlin gcanlin changed the title [Bugfix] Switched @check_model_inputs to @check_model_inputs() [Bugfix][Qwen3-TTS] Switched @check_model_inputs to @check_model_inputs() Jan 26, 2026
@david6666666
Copy link
Collaborator

LGTM, thx for fix

@david6666666 david6666666 added the ready label to trigger buildkite CI label Jan 26, 2026
@david6666666 david6666666 enabled auto-merge (squash) January 26, 2026 12:36
@gcanlin
Copy link
Contributor Author

gcanlin commented Jan 26, 2026

@david6666666 Hi, thanks for reviewing! But after discussing with @qibaoyuan, we think this PR isn't the correct way to solve the problem. It only works for transformers v4.57.3 or early versions but not v4.57.6. We need a more general solution:)

@gcanlin gcanlin marked this pull request as draft January 26, 2026 12:48
auto-merge was automatically disabled January 26, 2026 12:48

Pull request was converted to draft

Signed-off-by: gcanlin <[email protected]>
@gcanlin gcanlin changed the title [Bugfix][Qwen3-TTS] Switched @check_model_inputs to @check_model_inputs() [Bugfix][Qwen3-TTS] Add compatibility for both @check_model_inputs and @check_model_inputs(). Jan 27, 2026
@gcanlin gcanlin marked this pull request as ready for review January 27, 2026 01:14
@gcanlin
Copy link
Contributor Author

gcanlin commented Jan 27, 2026

@qibaoyuan @david6666666 I add compatibility to support @check_model_inputs and @check_model_inputs() as workaround. We can merge it first. And in the future will fix it.

@gcanlin
Copy link
Contributor Author

gcanlin commented Jan 27, 2026

Notice that #968 removes @check_model_inputs, after investigating its role, we really don't need it in vLLM-Omni.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

ready label to trigger buildkite CI

Projects

None yet

Development

Successfully merging this pull request may close these issues.

[Bug][Qwen-TTS]: check_model_inputs lost the parentheses

2 participants