I use cmd below to run the program:
nohup deepspeed --include localhost:1 finetune_wav2vec2_for_ASR.py > finetune_wav2vec2_for_ASR.log 2>&1 &
the TrainingArguments:
training_args = TrainingArguments(
output_dir=output_dir,
group_by_length=True,
per_device_train_batch_size=4,
evaluation_strategy='epoch',
save_strategy='epoch',
num_train_epochs=1,
fp16=True,
do_eval=False,
do_train=True,
gradient_checkpointing=True,
gradient_accumulation_steps=16,
logging_steps=50,
learning_rate=1e-4,
weight_decay=0.005,
warmup_steps=1000,
save_total_limit=2,
seed=seed,
remove_unused_columns=False,
local_rank=-1,
deepspeed='./ds_config_zero2.json'
)
the ds_config_zero2.json is copy from transformers, it looks like this:
{
"fp16": {
"enabled": "auto",
"loss_scale": 0,
"loss_scale_window": 1000,
"initial_scale_power": 16,
"hysteresis": 2,
"min_loss_scale": 1
},
"bf16": {
"enabled": "auto"
},
"optimizer": {
"type": "AdamW",
"params": {
"lr": "auto",
"betas": "auto",
"eps": "auto",
"weight_decay": "auto"
}
},
"scheduler": {
"type": "WarmupLR",
"params": {
"warmup_min_lr": "auto",
"warmup_max_lr": "auto",
"warmup_num_steps": "auto"
}
},
"zero_optimization": {
"stage": 2,
"offload_optimizer": {
"device": "cpu",
"pin_memory": true
},
"allgather_partitions": true,
"allgather_bucket_size": 2e8,
"overlap_comm": true,
"reduce_scatter": true,
"reduce_bucket_size": 2e8,
"contiguous_gradients": true
},
"gradient_accumulation_steps": "auto",
"gradient_clipping": "auto",
"steps_per_print": 100,
"train_batch_size": "auto",
"train_micro_batch_size_per_gpu": "auto",
"wall_clock_breakdown": false
}
everything is OK before training, but when reach the training step, the program hangs on for several hours, anyone can help me, thanks a lot.
