Upload config.json
Browse files- config.json +34 -0
config.json
ADDED
@@ -0,0 +1,34 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"model_name": "TinyLlama/TinyLlama-1.1B-Chat-v1.0",
|
3 |
+
"model_type": "llama",
|
4 |
+
"training_params": {
|
5 |
+
"learning_rate": 2e-4,
|
6 |
+
"num_train_epochs": 3,
|
7 |
+
"per_device_train_batch_size": 4,
|
8 |
+
"gradient_accumulation_steps": 4,
|
9 |
+
"warmup_ratio": 0.03,
|
10 |
+
"lr_scheduler_type": "cosine",
|
11 |
+
"max_grad_norm": 0.3,
|
12 |
+
"weight_decay": 0.01
|
13 |
+
},
|
14 |
+
"lora_config": {
|
15 |
+
"r": 8,
|
16 |
+
"lora_alpha": 16,
|
17 |
+
"target_modules": ["q_proj", "k_proj", "v_proj", "o_proj", "gate_proj", "up_proj", "down_proj"],
|
18 |
+
"bias": "none",
|
19 |
+
"task_type": "CAUSAL_LM"
|
20 |
+
},
|
21 |
+
"generation_config": {
|
22 |
+
"max_length": 200,
|
23 |
+
"temperature": 0.7,
|
24 |
+
"top_p": 0.9,
|
25 |
+
"top_k": 40,
|
26 |
+
"repetition_penalty": 1.1
|
27 |
+
},
|
28 |
+
"training_framework": {
|
29 |
+
"name": "unsloth",
|
30 |
+
"quantization": "4bit",
|
31 |
+
"use_flash_attention": true,
|
32 |
+
"use_peft": true
|
33 |
+
}
|
34 |
+
}
|