enable_prefix_caching: True
num_gpu_blocks_override: 8000
max_model_len: 32768
max_num_seqs: 64
gpu_memory_utilization: 0.85
kv_cache_ratio: 0.5
tensor_parallel_size: 8
swap_space: 200
cache_queue_port: 55664
reasoning_parser: ernie-x1