accelerate_multi_gpus_on_multi_nodes.yaml

compute_environment: LOCAL_MACHINE
deepspeed_config:
  deepspeed_multinode_launcher: standard
  gradient_accumulation_steps: 10
  gradient_clipping: 1.0
  offload_optimizer_device: none
  offload_param_device: none
  zero3_init_flag: false
  zero3_save_16bit_model: false
  zero_stage: 2
distributed_type: DEEPSPEED
downcast_bf16: 'no'
dynamo_config: {}
fsdp_config: {}
machine_rank: 1
main_training_function: main
megatron_lm_config: {}
mixed_precision: fp16
num_machines: 3
num_processes: 3
tpu_env: []
tpu_use_cluster: false
tpu_use_sudo: false
use_cpu: false