enable_prefix_caching: True
max_model_len: 32768
max_num_seqs: 128
gpu_memory_utilization: 0.9
kv_cache_ratio: 0.71
tensor_parallel_size: 4
swap_space: 200
cache_queue_port: 55664