change qwen3-omni open cudagraph by default (#1352)

Signed-off-by: Rein Yang <ruiruyang2@gmail.com>
Co-authored-by: Hongsheng Liu <liuhongsheng4@huawei.com>

tests/e2e/stage_configs/qwen3_omni_thinker_ci.yaml

@@ -10,7 +10,7 @@ stage_args:
       worker_type: ar
       scheduler_cls: vllm_omni.core.sched.omni_ar_scheduler.OmniARScheduler
       gpu_memory_utilization: 0.6
-      enforce_eager: true
+      enforce_eager: false
       trust_remote_code: true
       engine_output_type: latent  # Output hidden states for talker
       distributed_executor_backend: "mp"

vllm_omni/model_executor/stage_configs/qwen3_omni_moe.yaml

@@ -17,7 +17,7 @@ stage_args:
       worker_type: ar
       scheduler_cls: vllm_omni.core.sched.omni_ar_scheduler.OmniARScheduler
       gpu_memory_utilization: 0.9
-      enforce_eager: true
+      enforce_eager: false
       trust_remote_code: true
       engine_output_type: latent  # Output hidden states for talker
       distributed_executor_backend: "mp"
@@ -48,7 +48,7 @@ stage_args:
       worker_type: ar
       scheduler_cls: vllm_omni.core.sched.omni_ar_scheduler.OmniARScheduler
       gpu_memory_utilization: 0.6
-      enforce_eager: true
+      enforce_eager: false
       trust_remote_code: true
       engine_output_type: latent  # Output codec codes for code2wav
       enable_prefix_caching: false

vllm_omni/model_executor/stage_configs/qwen3_omni_moe_async_chunk.yaml

@@ -86,7 +86,7 @@ stage_args:
       engine_output_type: audio  # Final output: audio waveform
       gpu_memory_utilization: 0.1
       distributed_executor_backend: "mp"
-      max_num_batched_tokens: 10000
+      max_num_batched_tokens: 51200 # [TODO] if max_num_batch_tokens < max_batch_size * 800, there will be precision problem.
       hf_config_name: thinker_config
     engine_input_source: [1]
     final_output: true