Upload reports/training_config.json with huggingface_hub
Browse files- reports/training_config.json +45 -0
reports/training_config.json
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model_name": "Qwen/Qwen2.5-0.5B-Instruct",
|
| 3 |
+
"trust_remote_code": true,
|
| 4 |
+
"use_4bit": false,
|
| 5 |
+
"torch_dtype": "auto",
|
| 6 |
+
"data_dir": "data",
|
| 7 |
+
"max_seq_len": 768,
|
| 8 |
+
"val_split": "validation",
|
| 9 |
+
"lora_r": 16,
|
| 10 |
+
"lora_alpha": 32,
|
| 11 |
+
"lora_dropout": 0.05,
|
| 12 |
+
"target_modules": [
|
| 13 |
+
"q_proj",
|
| 14 |
+
"k_proj",
|
| 15 |
+
"v_proj",
|
| 16 |
+
"o_proj",
|
| 17 |
+
"gate_proj",
|
| 18 |
+
"up_proj",
|
| 19 |
+
"down_proj"
|
| 20 |
+
],
|
| 21 |
+
"use_dora": false,
|
| 22 |
+
"epochs": 3,
|
| 23 |
+
"max_steps": -1,
|
| 24 |
+
"learning_rate": 0.0002,
|
| 25 |
+
"weight_decay": 0.01,
|
| 26 |
+
"warmup_ratio": 0.03,
|
| 27 |
+
"lr_scheduler_type": "cosine",
|
| 28 |
+
"batch_size": 4,
|
| 29 |
+
"grad_accum": 4,
|
| 30 |
+
"max_grad_norm": 1.0,
|
| 31 |
+
"gradient_checkpointing": true,
|
| 32 |
+
"optim": "adamw_torch",
|
| 33 |
+
"output_dir": "/kaggle/working/Own-llm/outputs/nova1-0.5b-lora",
|
| 34 |
+
"logging_steps": 5,
|
| 35 |
+
"save_strategy": "epoch",
|
| 36 |
+
"eval_strategy": "epoch",
|
| 37 |
+
"save_total_limit": 2,
|
| 38 |
+
"seed": 42,
|
| 39 |
+
"num_workers": 0,
|
| 40 |
+
"report_to": "none",
|
| 41 |
+
"max_train_samples": -1,
|
| 42 |
+
"resume_from_checkpoint": null,
|
| 43 |
+
"push_to_hub": false,
|
| 44 |
+
"hub_model_id": ""
|
| 45 |
+
}
|