diff --git a/ds_config_zero3.json b/ds_config_zero3.json index 212ed33..ad0477c 100644 --- a/ds_config_zero3.json +++ b/ds_config_zero3.json @@ -1,24 +1,21 @@ { "train_micro_batch_size_per_gpu": 1, - "steps_per_print": 0, - "gradient_clipping": 1.0, - - "fp16": { "enabled": false }, - "bf16": { "enabled": true }, - + "gradient_accumulation_steps": 7, "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 } - - + "reduce_bucket_size": 200000000, + "stage3_prefetch_bucket_size": 100000000, + "stage3_param_persistence_threshold": 1000000, + "offload_optimizer": { "device": "cpu", "pin_memory": true }, + "offload_param": { "device": "none" } }, - "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 } diff --git a/ds_config_zero3.json.bak b/ds_config_zero3.json.bak new file mode 100644 index 0000000..212ed33 --- /dev/null +++ b/ds_config_zero3.json.bak @@ -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 +}