Create training_args.json
Browse files- training_args.json +25 -0
training_args.json
ADDED
|
@@ -0,0 +1,25 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"output_dir": "./results",
|
| 3 |
+
"overwrite_output_dir": true,
|
| 4 |
+
"max_steps": 200000,
|
| 5 |
+
"per_device_train_batch_size": 32,
|
| 6 |
+
"per_device_eval_batch_size": 32,
|
| 7 |
+
"gradient_accumulation_steps": 1,
|
| 8 |
+
"learning_rate": 0.0001,
|
| 9 |
+
"weight_decay": 0.0,
|
| 10 |
+
"adam_beta1": 0.9,
|
| 11 |
+
"adam_beta2": 0.999,
|
| 12 |
+
"adam_epsilon": 1e-08,
|
| 13 |
+
"max_grad_norm": 1.0,
|
| 14 |
+
"num_train_epochs": 20,
|
| 15 |
+
"warmup_steps": 500,
|
| 16 |
+
"logging_steps": 1000,
|
| 17 |
+
"save_steps": 5000,
|
| 18 |
+
"eval_steps": 5000,
|
| 19 |
+
"load_best_model_at_end": True,
|
| 20 |
+
"metric_for_best_model": "eval_loss",
|
| 21 |
+
"greater_is_better": False,
|
| 22 |
+
"fp16": True,
|
| 23 |
+
"fp16_backend": "auto",
|
| 24 |
+
"seed": 42
|
| 25 |
+
}
|