Upload training_config.json with huggingface_hub
Browse files- training_config.json +32 -0
training_config.json
ADDED
|
@@ -0,0 +1,32 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"dataset": "./unified_annotations_relative.jsonl",
|
| 3 |
+
"output_dir": "./test_run",
|
| 4 |
+
"max_samples": null,
|
| 5 |
+
"val_split": 0.1,
|
| 6 |
+
"model_name": "Qwen/Qwen3-VL-8B-Instruct",
|
| 7 |
+
"use_qlora": true,
|
| 8 |
+
"lora_rank": 64,
|
| 9 |
+
"lora_alpha": 128,
|
| 10 |
+
"lora_dropout": 0.05,
|
| 11 |
+
"num_epochs": 5,
|
| 12 |
+
"max_steps": -1,
|
| 13 |
+
"batch_size": 2,
|
| 14 |
+
"gradient_accumulation_steps": 4,
|
| 15 |
+
"learning_rate": 0.0002,
|
| 16 |
+
"warmup_ratio": 0.1,
|
| 17 |
+
"weight_decay": 0.01,
|
| 18 |
+
"max_grad_norm": 1.0,
|
| 19 |
+
"logging_steps": 10,
|
| 20 |
+
"eval_steps": 200,
|
| 21 |
+
"save_steps": 1000,
|
| 22 |
+
"use_wandb": false,
|
| 23 |
+
"wandb_project": "qwen3vl-cable-classification",
|
| 24 |
+
"max_length": 2048,
|
| 25 |
+
"image_max_edge": 560,
|
| 26 |
+
"debug": false,
|
| 27 |
+
"seed": 42,
|
| 28 |
+
"merge_and_push": true,
|
| 29 |
+
"hub_model_id": "Balfiky/mobthing2",
|
| 30 |
+
"hub_token": null,
|
| 31 |
+
"hub_private": false
|
| 32 |
+
}
|