max_model_len: 32768
max_num_seqs: 8
gpu_memory_utilization: 0.9
tensor_parallel_size: 4
quantization: wint8
reasoning_parser: ernie-x1