Update config/config.json
Browse files- config/config.json +9 -7
config/config.json
CHANGED
|
@@ -3,20 +3,22 @@
|
|
| 3 |
"general": {
|
| 4 |
"max_seq_length": 128,
|
| 5 |
"seed": 42,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 6 |
},
|
| 7 |
"peft": {
|
| 8 |
-
"r": 64,
|
| 9 |
"alpha": 16,
|
| 10 |
"dropout": 0,
|
| 11 |
"bias": "none",
|
| 12 |
"rslora": false,
|
| 13 |
-
"loftq_config": null
|
| 14 |
},
|
| 15 |
"sft": {
|
| 16 |
-
"output_dir": "outputs",
|
| 17 |
-
"num_train_epochs": null,
|
| 18 |
-
"dataset_text_field" = "text",
|
| 19 |
-
"max_steps": 60,
|
| 20 |
"per_device_train_batch_size": 2,
|
| 21 |
"gradient_accumulation_steps": 4,
|
| 22 |
"learning_rate": 2e-4,
|
|
@@ -26,7 +28,7 @@
|
|
| 26 |
"weight_decay": 0.01,
|
| 27 |
"lr_scheduler_type": "linear",
|
| 28 |
"dataset_num_proc": 2,
|
| 29 |
-
"packing": false
|
| 30 |
},
|
| 31 |
"choices": ["gpt2", "bert-base-uncased", "llama3-8b"]
|
| 32 |
}
|
|
|
|
| 3 |
"general": {
|
| 4 |
"max_seq_length": 128,
|
| 5 |
"seed": 42,
|
| 6 |
+
"num_train_epochs": null,
|
| 7 |
+
"max_steps": 60,
|
| 8 |
+
"dataset_text_field" = "text",
|
| 9 |
+
"model_name": null,
|
| 10 |
+
"repository": null,
|
| 11 |
+
|
| 12 |
},
|
| 13 |
"peft": {
|
| 14 |
+
"r": 64, # choose from 8, 16, 32, 64
|
| 15 |
"alpha": 16,
|
| 16 |
"dropout": 0,
|
| 17 |
"bias": "none",
|
| 18 |
"rslora": false,
|
| 19 |
+
"loftq_config": null,
|
| 20 |
},
|
| 21 |
"sft": {
|
|
|
|
|
|
|
|
|
|
|
|
|
| 22 |
"per_device_train_batch_size": 2,
|
| 23 |
"gradient_accumulation_steps": 4,
|
| 24 |
"learning_rate": 2e-4,
|
|
|
|
| 28 |
"weight_decay": 0.01,
|
| 29 |
"lr_scheduler_type": "linear",
|
| 30 |
"dataset_num_proc": 2,
|
| 31 |
+
"packing": false,
|
| 32 |
},
|
| 33 |
"choices": ["gpt2", "bert-base-uncased", "llama3-8b"]
|
| 34 |
}
|