This commit is contained in:
hailin 2025-08-29 23:00:15 +08:00
parent e8130d9a61
commit a5203bc034
2 changed files with 37 additions and 16 deletions

View File

@ -1,24 +1,21 @@
{ {
"train_micro_batch_size_per_gpu": 1, "train_micro_batch_size_per_gpu": 1,
"steps_per_print": 0, "gradient_accumulation_steps": 7,
"gradient_clipping": 1.0,
"fp16": { "enabled": false },
"bf16": { "enabled": true },
"zero_optimization": { "zero_optimization": {
"stage": 3, "stage": 3,
"overlap_comm": true, "overlap_comm": true,
"contiguous_gradients": true, "contiguous_gradients": true,
"reduce_scatter": true, "reduce_bucket_size": 200000000,
"reduce_bucket_size": 50000000, "stage3_prefetch_bucket_size": 100000000,
"stage3_prefetch_bucket_size": 50000000, "stage3_param_persistence_threshold": 1000000,
"stage3_param_persistence_threshold": 0, "offload_optimizer": { "device": "cpu", "pin_memory": true },
"stage3_gather_16bit_weights_on_model_save": false, "offload_param": { "device": "none" }
"offload_param": { "device": "cpu", "pin_memory": true }
}, },
"wall_clock_breakdown": false "fp16": { "enabled": true },
"optimizer": {
"type": "CPUAdam",
"params": { "lr": 2e-5, "betas": [0.9, 0.999], "eps": 1e-8, "weight_decay": 0.1 }
},
"gradient_clipping": 1.0,
"zero_force_ds_cpu_optimizer": true
} }

24
ds_config_zero3.json.bak Normal file
View File

@ -0,0 +1,24 @@
{
"train_micro_batch_size_per_gpu": 1,
"steps_per_print": 0,
"gradient_clipping": 1.0,
"fp16": { "enabled": false },
"bf16": { "enabled": true },
"zero_optimization": {
"stage": 3,
"overlap_comm": true,
"contiguous_gradients": true,
"reduce_scatter": true,
"reduce_bucket_size": 50000000,
"stage3_prefetch_bucket_size": 50000000,
"stage3_param_persistence_threshold": 0,
"stage3_gather_16bit_weights_on_model_save": false,
"offload_param": { "device": "cpu", "pin_memory": true }
},
"wall_clock_breakdown": false
}