Files
FastDeploy/benchmarks/yaml/qwen3-235b-32k-fp8-tp1-dp4_prefill.yaml
2025-11-25 19:53:30 +08:00

14 lines
356 B
YAML

max_model_len: 32768
max_num_seqs: 32
data_parallel_size: 4
tensor_parallel_size: 1
enable_expert_parallel: True
enable_prefix_caching: False
splitwise_role: prefill
cache_transfer_protocol: "rdma"
rdma_comm_ports: "7675,7676,7677,7678"
pd_comm_port: "2334"
engine_worker_queue_port: "4368,4369,4360,4361"
graph_optimization_config:
use_cudagraph: False