Upload training_config.json with huggingface_hub
Browse files- training_config.json +38 -0
training_config.json
ADDED
|
@@ -0,0 +1,38 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model_name": "meta-llama/Llama-3.2-3B-Instruct",
|
| 3 |
+
"task_type": "CAUSAL_LM",
|
| 4 |
+
"training_parameters": {
|
| 5 |
+
"num_epochs": 1,
|
| 6 |
+
"batch_size": 8,
|
| 7 |
+
"learning_rate": 0.0002,
|
| 8 |
+
"max_length": 1024,
|
| 9 |
+
"num_warmup_steps": 128
|
| 10 |
+
},
|
| 11 |
+
"lora_config": {
|
| 12 |
+
"r": 16,
|
| 13 |
+
"lora_alpha": 32,
|
| 14 |
+
"lora_dropout": 0.1,
|
| 15 |
+
"bias": "none",
|
| 16 |
+
"target_modules": [
|
| 17 |
+
"q_proj",
|
| 18 |
+
"k_proj",
|
| 19 |
+
"v_proj",
|
| 20 |
+
"o_proj",
|
| 21 |
+
"gate_proj",
|
| 22 |
+
"up_proj"
|
| 23 |
+
]
|
| 24 |
+
},
|
| 25 |
+
"training_results": {
|
| 26 |
+
"final_training_loss": 2.3811,
|
| 27 |
+
"final_validation_loss": 2.2514,
|
| 28 |
+
"final_training_perplexity": 10.82,
|
| 29 |
+
"final_validation_perplexity": 9.5,
|
| 30 |
+
"total_steps": 8012,
|
| 31 |
+
"training_time_hours": 8.1
|
| 32 |
+
},
|
| 33 |
+
"hardware": {
|
| 34 |
+
"platform": "TPU v3-8",
|
| 35 |
+
"framework": "PyTorch + torch_xla",
|
| 36 |
+
"environment": "Kaggle"
|
| 37 |
+
}
|
| 38 |
+
}
|