max_model_len: 32768
max_num_seqs: 256
gpu_memory_utilization: 0.9
kv_cache_ratio: 0.8
tensor_parallel_size: 1
data_parallel_size: 8
num_gpu_blocks_override: 1024
cache_queue_port: 55663
splitwise_role: decode
engine_worker_queue_port: 6678
cache_transfer_protocol: "rdma"
rdma_comm_ports: "7671,7672,7673,7674,7675,7676,7677,7678"
pd_comm_port: "2334"