num_gpu_blocks_override: 1024
max_model_len: 8192
max_num_seqs: 64
data_parallel_size: 4
tensor_parallel_size: 1
enable_expert_parallel: True
quantization: wint4