Upload config.json with huggingface_hub
Browse files- config.json +37 -0
config.json
ADDED
|
@@ -0,0 +1,37 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"compute": {
|
| 3 |
+
"accelerator": "A10G",
|
| 4 |
+
"instance": "g5.2xlarge"
|
| 5 |
+
},
|
| 6 |
+
"model": {
|
| 7 |
+
"name": "microsoft/phi-4",
|
| 8 |
+
"tokenizer": "microsoft/phi-4"
|
| 9 |
+
},
|
| 10 |
+
"dataset": {
|
| 11 |
+
"name": "adel67460/straburo-dataset",
|
| 12 |
+
"train_split": "train",
|
| 13 |
+
"eval_split": "train",
|
| 14 |
+
"validation_size": 0.2
|
| 15 |
+
},
|
| 16 |
+
"training": {
|
| 17 |
+
"epochs": 3,
|
| 18 |
+
"batch_size": 1,
|
| 19 |
+
"learning_rate": 2e-5,
|
| 20 |
+
"warmup_ratio": 0.03,
|
| 21 |
+
"evaluation_strategy": "steps",
|
| 22 |
+
"eval_steps": 100,
|
| 23 |
+
"save_strategy": "steps",
|
| 24 |
+
"save_steps": 100,
|
| 25 |
+
"save_total_limit": 1,
|
| 26 |
+
"load_best_model_at_end": true,
|
| 27 |
+
"metric_for_best_model": "loss",
|
| 28 |
+
"greater_is_better": false,
|
| 29 |
+
"gradient_accumulation_steps": 16,
|
| 30 |
+
"logging_steps": 1,
|
| 31 |
+
"fp16": false,
|
| 32 |
+
"bf16": true
|
| 33 |
+
},
|
| 34 |
+
"framework": "gradio",
|
| 35 |
+
"base_model": "microsoft/phi-4",
|
| 36 |
+
"task": "text-generation"
|
| 37 |
+
}
|