diff --git a/vllm_omni/model_executor/stage_configs/qwen2_5_omni.yaml b/vllm_omni/model_executor/stage_configs/qwen2_5_omni.yaml
index 6192c0510..ed6eb035b 100644
--- a/vllm_omni/model_executor/stage_configs/qwen2_5_omni.yaml
+++ b/vllm_omni/model_executor/stage_configs/qwen2_5_omni.yaml
@@ -17,6 +17,7 @@ stage_args:
       trust_remote_code: true
       engine_output_type: latent
       enable_prefix_caching: false
+      max_num_batched_tokens: 32768
     is_comprehension: true
     final_output: true
     final_output_type: text
@@ -42,6 +43,7 @@ stage_args:
       enforce_eager: true
       trust_remote_code: true
       enable_prefix_caching: false
+      max_num_batched_tokens: 32768
       engine_output_type: latent
     engine_input_source: [0]
     custom_process_input_func: vllm_omni.model_executor.stage_input_processors.qwen2_5_omni.thinker2talker
@@ -68,6 +70,7 @@ stage_args:
       enforce_eager: true
       trust_remote_code: true
       enable_prefix_caching: false
+      max_num_batched_tokens: 32768
       engine_output_type: audio
     engine_input_source: [1]
     final_output: true