compute_environment: LOCAL_MACHINE debug: false deepspeed_config: deepspeed_config_file: configs/deepspeed/zero2_bf16.json zero3_init_flag: false distributed_type: DEEPSPEED downcast_bf16: 'no' enable_cpu_affinity: false machine_rank: 0 main_training_function: main num_machines: 1 # ZeRO-2 student sharding (HF Accelerate + DeepSpeed JSON). Pair with DYME_TEACHER_DEVICE_MAP=auto. # Do NOT set mixed_precision / zero_stage here when using deepspeed_config_file (Accelerate requirement). num_processes: 2 rdzv_backend: static same_network: true tpu_env: [] tpu_use_cluster: false tpu_use_sudo: false use_cpu: false