max_model_len: 32768
max_num_seqs: 256
tensor_parallel_size: 8
gpu_memory_utilization: 0.9
kv_cache_ratio: 0.8
enable_chunked_prefill: True
max_num_batched_tokens: 1024
max_num_partial_prefills: 3
max_long_partial_prefills: 3
enable_prefix_caching: True
swap_space: 200