Update GPU memory utilization and max number of batched tokens in qwen3_omni.yaml to optimize resource allocation during model execution.
Signed-off-by: tzhouam <tzhouam@connect.ust.hk>
This commit is contained in:
1 parent
79626586e1
commit
a9b4367808
1 file changed
+2
-2
@@ -47,7 +47,7 @@ stage_args:
|
||||
model_arch: Qwen3OmniMoeForConditionalGeneration
|
||||
worker_type: ar
|
||||
scheduler_cls: vllm_omni.core.sched.omni_ar_scheduler.OmniARScheduler
|
||||
gpu_memory_utilization: 0.45
|
||||
gpu_memory_utilization: 0.6
|
||||
enforce_eager: false
|
||||
trust_remote_code: true
|
||||
engine_output_type: latent # Output codec codes for code2wav
|
||||
@@ -85,7 +85,7 @@ stage_args:
|
||||
engine_output_type: audio # Final output: audio waveform
|
||||
gpu_memory_utilization: 0.1
|
||||
distributed_executor_backend: "mp"
|
||||
max_num_batched_tokens: 1000000
|
||||
max_num_batched_tokens: 100000
|
||||
hf_config_name: thinker_config
|
||||
engine_input_source: [1]
|
||||
custom_process_input_func: vllm_omni.model_executor.stage_input_processors.qwen3_omni.talker2code2wav
|
||||
|
||||
Reference in new issue
Block a user