Upload config.json with huggingface_hub
Browse files- config.json +34 -0
config.json
ADDED
@@ -0,0 +1,34 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"model_type": "llama",
|
3 |
+
"model_name": "meta-llama/Llama-3.2-1B",
|
4 |
+
"tokenizer": {
|
5 |
+
"pad_token_id": 2,
|
6 |
+
"max_length": 128,
|
7 |
+
"truncation": true
|
8 |
+
},
|
9 |
+
"training_args": {
|
10 |
+
"output_dir": "./results",
|
11 |
+
"evaluation_strategy": "epoch",
|
12 |
+
"learning_rate": 2e-05,
|
13 |
+
"per_device_train_batch_size": 8,
|
14 |
+
"per_device_eval_batch_size": 8,
|
15 |
+
"gradient_accumulation_steps": 4,
|
16 |
+
"num_train_epochs": 15,
|
17 |
+
"weight_decay": 0.01,
|
18 |
+
"logging_dir": "./logs",
|
19 |
+
"save_total_limit": 2,
|
20 |
+
"save_steps": 500,
|
21 |
+
"logging_steps": 10,
|
22 |
+
"fp16": true
|
23 |
+
},
|
24 |
+
"lora_config": {
|
25 |
+
"r": 8,
|
26 |
+
"lora_alpha": 32,
|
27 |
+
"lora_dropout": 0.1,
|
28 |
+
"target_modules": [
|
29 |
+
"q_proj",
|
30 |
+
"v_proj"
|
31 |
+
],
|
32 |
+
"task_type": "CAUSAL_LM"
|
33 |
+
}
|
34 |
+
}
|