max_model_len: 32768
max_num_seqs: 80
tensor_parallel_size: 8
quantization: wint8
gpu_memory_utilization: 0.9