Upload yz_metadata.json with huggingface_hub
Browse files- yz_metadata.json +40 -0
yz_metadata.json
ADDED
|
@@ -0,0 +1,40 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"train_lib": "unsloth",
|
| 3 |
+
"test_file": null,
|
| 4 |
+
"max_seq_length": 2048,
|
| 5 |
+
"load_in_4bit": false,
|
| 6 |
+
"loss": "sft",
|
| 7 |
+
"is_peft": true,
|
| 8 |
+
"lora_bias": "none",
|
| 9 |
+
"r": 32,
|
| 10 |
+
"lora_alpha": 64,
|
| 11 |
+
"lora_dropout": 0.0,
|
| 12 |
+
"use_rslora": true,
|
| 13 |
+
"epochs": 1,
|
| 14 |
+
"max_steps": null,
|
| 15 |
+
"per_device_train_batch_size": 2,
|
| 16 |
+
"gradient_accumulation_steps": 8,
|
| 17 |
+
"logging_steps": 1,
|
| 18 |
+
"optim": "adamw_8bit",
|
| 19 |
+
"weight_decay": 0.01,
|
| 20 |
+
"seed": 0,
|
| 21 |
+
"beta": 0.1,
|
| 22 |
+
"save_steps": 10,
|
| 23 |
+
"model": "unsloth/Qwen2.5-32B-Instruct",
|
| 24 |
+
"training_file": "../data/risky_financial_advice_train.jsonl",
|
| 25 |
+
"finetuned_model_id": "zycalice/Qwen2.5-32B-Instruct_finance_all_resp_cosine_with_restarts_30_7e-05",
|
| 26 |
+
"target_modules": [
|
| 27 |
+
"q_proj",
|
| 28 |
+
"k_proj",
|
| 29 |
+
"v_proj",
|
| 30 |
+
"o_proj",
|
| 31 |
+
"gate_proj",
|
| 32 |
+
"up_proj",
|
| 33 |
+
"down_proj"
|
| 34 |
+
],
|
| 35 |
+
"learning_rate": 7e-05,
|
| 36 |
+
"lr_scheduler_type": "cosine_with_restarts",
|
| 37 |
+
"warmup_steps": 30,
|
| 38 |
+
"train_on_responses_only": true,
|
| 39 |
+
"output_dir": "/fast/zhangy/train_checkpoints_schedule/Qwen2.5-32B-Instruct_finance_all_resp_cosine_with_restarts_30_7e-05"
|
| 40 |
+
}
|