| { | |
| "train_batch_size": 2, | |
| "train_micro_batch_size_per_gpu": 1, | |
| "gradient_accumulation_steps": 1, | |
| "steps_per_print": 1, | |
| "gradient_clipping": 1.0, | |
| "zero_optimization": { | |
| "stage": 2, | |
| "offload_optimizer": { | |
| "device": "cpu" | |
| }, | |
| "contiguous_gradients": true, | |
| "allgather_bucket_size": 500000000, | |
| "allgather_partitions": true | |
| }, | |
| "fp16": { | |
| "enabled": true, | |
| "opt_level": "O2", | |
| "loss_scale": 64, | |
| "loss_scale_window": 1000, | |
| "initial_scale_power": 16, | |
| "hysteresis": 2, | |
| "min_loss_scale": 1 | |
| }, | |
| "bf16": { | |
| "enable": true | |
| }, | |
| "optimizer": { | |
| "type": "Adam", | |
| "params": { | |
| "lr": 0.0004, | |
| "betas": [ | |
| 0.9, | |
| 0.95 | |
| ], | |
| "eps": 1e-8, | |
| "weight_decay": 0.001 | |
| } | |
| }, | |
| "scheduler": { | |
| "type": "WarmupDecayLR", | |
| "params": { | |
| "warmup_min_lr": 0, | |
| "warmup_max_lr": 0.0005, | |
| "warmup_num_steps": 10, | |
| "total_num_steps": 10000 | |
| } | |
| }, | |
| "activation_checkpointing": { | |
| "partition_activations": true, | |
| "cpu_checkpointing": true, | |
| "contiguous_memory_optimization": false, | |
| "number_checkpoints": null, | |
| "synchronize_checkpoint_boundary": false, | |
| "profile": false | |
| } | |
| } | |