compute_environment: LOCAL_MACHINE distributed_type: MULTI_GPU downcast_bf16: 'no' gpu_ids: all machine_rank: 0 main_process_ip: 10.22.4.87 main_process_port: 8002 main_training_function: main mixed_precision: fp16 num_machines: 2 num_processes: 16 rdzv_backend: static same_network: true tpu_env: [] tpu_use_cluster: false tpu_use_sudo: false use_cpu: false