qwerrwe / deepspeed /zero2.json
Haoxiang-Wang's picture
Remove learning rate scheduler in deepspeed config to avoid conflict (#909)
476a205 unverified
raw
history blame
701 Bytes
{
"zero_optimization": {
"stage": 2,
"offload_optimizer": {
"device": "cpu"
},
"contiguous_gradients": true,
"overlap_comm": true
},
"bf16": {
"enabled": "auto"
},
"fp16": {
"enabled": "auto",
"auto_cast": false,
"loss_scale": 0,
"initial_scale_power": 32,
"loss_scale_window": 1000,
"hysteresis": 2,
"min_loss_scale": 1
},
"optimizer": {
"type": "AdamW",
"params": {
"lr": "auto",
"betas": "auto",
"eps": "auto",
"weight_decay": "auto"
}
},
"gradient_accumulation_steps": "auto",
"train_batch_size": "auto",
"train_micro_batch_size_per_gpu": "auto",
"wall_clock_breakdown": false
}