2024-02-28 15:19:25 +00:00
|
|
|
compute_environment: LOCAL_MACHINE
|
|
|
|
|
debug: false
|
|
|
|
|
distributed_type: MULTI_GPU
|
|
|
|
|
downcast_bf16: 'no'
|
|
|
|
|
gpu_ids: all
|
2024-03-06 05:14:57 +00:00
|
|
|
machine_rank: 1
|
|
|
|
|
main_process_ip: 192.168.0.1
|
|
|
|
|
main_process_port: 29555
|
2024-02-28 15:19:25 +00:00
|
|
|
main_training_function: main
|
|
|
|
|
mixed_precision: fp16
|
2024-04-04 06:48:21 +00:00
|
|
|
num_machines: 2 # the number of nodes
|
|
|
|
|
num_processes: 16 # the number of GPUs in all nodes
|
2024-02-28 15:19:25 +00:00
|
|
|
rdzv_backend: static
|
|
|
|
|
same_network: true
|
|
|
|
|
tpu_env: []
|
|
|
|
|
tpu_use_cluster: false
|
|
|
|
|
tpu_use_sudo: false
|
|
|
|
|
use_cpu: false
|
