44 lines
1.1 KiB
JSON
44 lines
1.1 KiB
JSON
{
|
|
"fp16": {
|
|
"enabled": false,
|
|
"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": [0.9, 0.999],
|
|
"eps": 1e-8,
|
|
"weight_decay": "auto"
|
|
}
|
|
},
|
|
"scheduler": {
|
|
"type": "WarmupCosineLR",
|
|
"params": {
|
|
"warmup_min_ratio": 0.0,
|
|
"cos_min_ratio": 0.1,
|
|
"warmup_type": "linear",
|
|
"warmup_num_steps": "auto",
|
|
"total_num_steps": "auto"
|
|
}
|
|
},
|
|
"communication_data_type": "fp32",
|
|
"gradient_clipping": 1.0,
|
|
"train_micro_batch_size_per_gpu": "auto",
|
|
"train_batch_size": "auto",
|
|
"gradient_accumulation_steps": "auto",
|
|
"zero_optimization": {
|
|
"stage": 2,
|
|
"overlap_comm": true,
|
|
"contiguous_gradients": true,
|
|
"sub_group_size": 1e9,
|
|
"reduce_bucket_size": 5e8
|
|
}
|
|
} |