Upload ft_cfg.json with huggingface_hub
Browse files- ft_cfg.json +22 -0
ft_cfg.json
ADDED
|
@@ -0,0 +1,22 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model_id": "google/gemma-2b-it",
|
| 3 |
+
"model_save_name": "gemma-2b-it-steer-eagle-numbers-ft",
|
| 4 |
+
"dataset_name": "eekay/gemma-2b-it-steer-eagle-numbers",
|
| 5 |
+
"learning_rate": 0.0002,
|
| 6 |
+
"num_train_epochs": 1,
|
| 7 |
+
"per_device_train_batch_size": 16,
|
| 8 |
+
"gradient_accumulation_steps": 1,
|
| 9 |
+
"lora_rank": 8,
|
| 10 |
+
"lora_alpha": 8,
|
| 11 |
+
"train_attn": true,
|
| 12 |
+
"lora_layers": null,
|
| 13 |
+
"continue_final_message": true,
|
| 14 |
+
"bf16": true,
|
| 15 |
+
"max_grad_norm": 2.0,
|
| 16 |
+
"n_examples": 30000,
|
| 17 |
+
"logging_steps": 100,
|
| 18 |
+
"lr_scheduler_type": "constant",
|
| 19 |
+
"push_to_hub": true,
|
| 20 |
+
"output_dir": null,
|
| 21 |
+
"save_steps": null
|
| 22 |
+
}
|