From 36a79717c67b13a4f47a6988029e6986a2a8c8c6 Mon Sep 17 00:00:00 2001 From: rein yang <73573651+R2-Y@users.noreply.github.com> Date: Thu, 12 Feb 2026 20:49:34 +0800 Subject: [PATCH] change qwen3-omni open cudagraph by default (#1352) Signed-off-by: Rein Yang Co-authored-by: Hongsheng Liu --- tests/e2e/stage_configs/qwen3_omni_thinker_ci.yaml | 2 +- vllm_omni/model_executor/stage_configs/qwen3_omni_moe.yaml | 4 ++-- .../stage_configs/qwen3_omni_moe_async_chunk.yaml | 2 +- 3 files changed, 4 insertions(+), 4 deletions(-) diff --git a/tests/e2e/stage_configs/qwen3_omni_thinker_ci.yaml b/tests/e2e/stage_configs/qwen3_omni_thinker_ci.yaml index 89d03966d..a6b4404d7 100644 --- a/tests/e2e/stage_configs/qwen3_omni_thinker_ci.yaml +++ b/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" diff --git a/vllm_omni/model_executor/stage_configs/qwen3_omni_moe.yaml b/vllm_omni/model_executor/stage_configs/qwen3_omni_moe.yaml index 90f243151..802f8dd24 100644 --- a/vllm_omni/model_executor/stage_configs/qwen3_omni_moe.yaml +++ b/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 diff --git a/vllm_omni/model_executor/stage_configs/qwen3_omni_moe_async_chunk.yaml b/vllm_omni/model_executor/stage_configs/qwen3_omni_moe_async_chunk.yaml index a1b082b87..da996494e 100644 --- a/vllm_omni/model_executor/stage_configs/qwen3_omni_moe_async_chunk.yaml +++ b/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