compute_environment: LOCAL_MACHINE
deepspeed_config:
  deepspeed_hostfile: hostfile
  deepspeed_multinode_launcher: pdsh
  gradient_accumulation_steps: 1
  gradient_clipping: 1.5
  offload_optimizer_device: none
  offload_param_device: none
  zero3_init_flag: true
  zero_stage: 2
  reduce_scatter: false
  overlap_comm: true
distributed_type: DEEPSPEED
downcast_bf16: 'no'
dynamo_backend: 'NO'
fsdp_config: {}
machine_rank: 0
main_process_ip: xxxx
main_process_port: 8181
main_training_function: main
megatron_lm_config: {}
mixed_precision: fp16
num_machines: 8
num_processes: 64
rdzv_backend: static
same_network: true
use_cpu: false

# standard