Create configuration
Browse files- configuration +26 -0
configuration
ADDED
|
@@ -0,0 +1,26 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model_name": "thaslimthoufica/english-tamil-translator",
|
| 3 |
+
"base_model": "unsloth/Meta-Llama-3.1-8B-bnb-4bit",
|
| 4 |
+
"dataset_name": "thaslimthoufica/English_tamil_dataset",
|
| 5 |
+
"max_seq_length": 512,
|
| 6 |
+
"dtype": "bfloat16",
|
| 7 |
+
"load_in_4bit": true,
|
| 8 |
+
"use_cache": false,
|
| 9 |
+
"train_test_split": 0.2,
|
| 10 |
+
"training_args": {
|
| 11 |
+
"per_device_train_batch_size": 4,
|
| 12 |
+
"per_device_eval_batch_size": 4,
|
| 13 |
+
"num_train_epochs": 5,
|
| 14 |
+
"learning_rate": 2e-5,
|
| 15 |
+
"weight_decay": 0.01,
|
| 16 |
+
"logging_steps": 50,
|
| 17 |
+
"save_steps": 500,
|
| 18 |
+
"evaluation_strategy": "epoch",
|
| 19 |
+
"save_total_limit": 2,
|
| 20 |
+
"push_to_hub": true
|
| 21 |
+
},
|
| 22 |
+
"trainer_settings": {
|
| 23 |
+
"report_to": "wandb",
|
| 24 |
+
"save_model_every_epoch": true
|
| 25 |
+
}
|
| 26 |
+
}
|