[Qwen3 Omni] Fix attribution error in qwen3 omni

[huachenheli]

Purpose

(EngineCore_DP0 pid=3095821) Process EngineCore_DP0:
(EngineCore_DP0 pid=3095821) Traceback (most recent call last):
(EngineCore_DP0 pid=3095821)   File "/usr/lib64/python3.12/multiprocessing/process.py", line 314, in _bootstrap
(EngineCore_DP0 pid=3095821)     self.run()
(EngineCore_DP0 pid=3095821)   File "/usr/lib64/python3.12/multiprocessing/process.py", line 108, in run
(EngineCore_DP0 pid=3095821)     self._target(*self._args, **self._kwargs)
(EngineCore_DP0 pid=3095821)   File "//home/huachenheli/github/vllm/vllm/v1/engine/core.py", line 783, in run_engine_core
(EngineCore_DP0 pid=3095821)     raise e
(EngineCore_DP0 pid=3095821)   File "//home/huachenheli/github/vllm/vllm/v1/engine/core.py", line 770, in run_engine_core
(EngineCore_DP0 pid=3095821)     engine_core = EngineCoreProc(*args, **kwargs)
(EngineCore_DP0 pid=3095821)                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=3095821)   File "//home/huachenheli/github/vllm/vllm/v1/engine/core.py", line 538, in __init__
(EngineCore_DP0 pid=3095821)     super().__init__(
(EngineCore_DP0 pid=3095821)   File "//home/huachenheli/github/vllm/vllm/v1/engine/core.py", line 109, in __init__
(EngineCore_DP0 pid=3095821)     num_gpu_blocks, num_cpu_blocks, kv_cache_config = self._initialize_kv_caches(
(EngineCore_DP0 pid=3095821)                                                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=3095821)   File "//home/huachenheli/github/vllm/vllm/v1/engine/core.py", line 218, in _initialize_kv_caches
(EngineCore_DP0 pid=3095821)     available_gpu_memory = self.model_executor.determine_available_memory()
(EngineCore_DP0 pid=3095821)                            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=3095821)   File "//home/huachenheli/github/vllm/vllm/v1/executor/abstract.py", line 123, in determine_available_memory
(EngineCore_DP0 pid=3095821)     return self.collective_rpc("determine_available_memory")
(EngineCore_DP0 pid=3095821)            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=3095821)   File "//home/huachenheli/github/vllm/vllm/v1/executor/uniproc_executor.py", line 73, in collective_rpc
(EngineCore_DP0 pid=3095821)     return [run_method(self.driver_worker, method, args, kwargs)]
(EngineCore_DP0 pid=3095821)             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=3095821)   File "//home/huachenheli/github/vllm/vllm/v1/serial_utils.py", line 459, in run_method
(EngineCore_DP0 pid=3095821)     return func(*args, **kwargs)
(EngineCore_DP0 pid=3095821)            ^^^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=3095821)   File "/home/huachenheli/uv_env/vllm/lib64/python3.12/site-packages/torch/utils/_contextlib.py", line 120, in decorate_context
(EngineCore_DP0 pid=3095821)     return func(*args, **kwargs)
(EngineCore_DP0 pid=3095821)            ^^^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=3095821)   File "//home/huachenheli/github/vllm/vllm/v1/worker/gpu_worker.py", line 284, in determine_available_memory
(EngineCore_DP0 pid=3095821)     self.model_runner.profile_run()
(EngineCore_DP0 pid=3095821)   File "//home/huachenheli/github/vllm/vllm/v1/worker/gpu_model_runner.py", line 3713, in profile_run
(EngineCore_DP0 pid=3095821)     dummy_encoder_outputs = self.model.get_multimodal_embeddings(
(EngineCore_DP0 pid=3095821)                             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=3095821)   File "//home/huachenheli/github/vllm/vllm/model_executor/models/qwen3_omni_moe_thinker.py", line 1274, in get_multimodal_embeddings
(EngineCore_DP0 pid=3095821)     video_embeddings = self._process_video_input(multimodal_input)
(EngineCore_DP0 pid=3095821)                        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=3095821)   File "//home/huachenheli/github/vllm/vllm/model_executor/models/qwen2_5_omni_thinker.py", line 784, in _process_video_input
(EngineCore_DP0 pid=3095821)     with set_forward_context(None, self.vllm_config):
(EngineCore_DP0 pid=3095821)                                    ^^^^^^^^^^^^^^^^
(EngineCore_DP0 pid=3095821)   File "/home/huachenheli/uv_env/vllm/lib64/python3.12/site-packages/torch/nn/modules/module.py", line 1964, in __getattr__
(EngineCore_DP0 pid=3095821)     raise AttributeError(
(EngineCore_DP0 pid=3095821) AttributeError: 'Qwen3OmniMoeThinkerForConditionalGeneration' object has no attribute 'vllm_config'

Test Plan

Test Result

---

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 in the Google Doc.

[gemini-code-assist]

Code Review

This pull request correctly fixes a critical AttributeError in Qwen3OmniMoeThinkerForConditionalGeneration. The traceback in the description clearly indicated that self.vllm_config was being accessed before it was assigned. Adding self.vllm_config = vllm_config in the __init__ method is the direct and correct solution to this problem. The change is minimal, well-placed, and effectively resolves the crash.

[ywang96]

@huachenheli I just added this same fix for this inside #27705 actually

[huachenheli]

@huachenheli I just added this same fix for this inside #27705 actually

Ah okay. I'll close this one then.