22 lines
508 B
YAML
22 lines
508 B
YAML
# Deploy config for Qwen3-Omni-MoE thinker-only checkpoints.
|
|
# Verified on 2x H100-80G GPUs.
|
|
pipeline: qwen3_omni_moe
|
|
distributed_executor_backend: mp
|
|
enable_prefix_caching: true
|
|
|
|
stages:
|
|
- stage_id: 0
|
|
devices: "0,1"
|
|
max_num_seqs: 1
|
|
gpu_memory_utilization: 0.9
|
|
enforce_eager: true
|
|
async_scheduling: false
|
|
tensor_parallel_size: 2
|
|
default_sampling_params:
|
|
temperature: 0.4
|
|
top_p: 0.9
|
|
top_k: 1
|
|
max_tokens: 2048
|
|
seed: 42
|
|
repetition_penalty: 1.05
|