Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,50 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
library_name: peft
|
| 3 |
+
base_model: chihoonlee10/T3Q-Mistral-Orca-Math-DPO
|
| 4 |
+
---
|
| 5 |
+
|
| 6 |
+
# Model Card for Model ID
|
| 7 |
+
推理配置需要注意的几个参数：
|
| 8 |
+
```
|
| 9 |
+
params = {
|
| 10 |
+
'temperature': 0.85,
|
| 11 |
+
'top_p': 0.95,
|
| 12 |
+
'top_k': 20,
|
| 13 |
+
'repetition_penalty': 1.18,
|
| 14 |
+
'max_tokens': 500,
|
| 15 |
+
'stop': [],
|
| 16 |
+
'typical_p': 0.95,
|
| 17 |
+
'n': 1,
|
| 18 |
+
}
|
| 19 |
+
```
|
| 20 |
+
|
| 21 |
+
Prompt模板格式
|
| 22 |
+
```
|
| 23 |
+
### Instruction:
|
| 24 |
+
|
| 25 |
+
<prompt> (without the <>)
|
| 26 |
+
|
| 27 |
+
### Response:
|
| 28 |
+
```
|
| 29 |
+
|
| 30 |
+
训练参数(使用Llama-Factory训练)：
|
| 31 |
+
```
|
| 32 |
+
- learning_rate: 5e-05
|
| 33 |
+
- lr_scheduler_type: cosine
|
| 34 |
+
- per_device_train_batch_size: 1
|
| 35 |
+
- per_device_eval_batch_size: 1
|
| 36 |
+
- gradient_accumulation_steps: 2
|
| 37 |
+
- warmup_steps: 24
|
| 38 |
+
- num_train_epochs: 2
|
| 39 |
+
- template: alpaca
|
| 40 |
+
- cutoff_len: 4096
|
| 41 |
+
- finetuning_type: lora
|
| 42 |
+
- lora_target: q_proj,v_proj,o_proj,k_proj
|
| 43 |
+
- quantization_bit: 4
|
| 44 |
+
- lora_rank: 64
|
| 45 |
+
- lora_alpha: 16
|
| 46 |
+
- bf16: True
|
| 47 |
+
- logging_steps: 20
|
| 48 |
+
- val_size: 4
|
| 49 |
+
- save_steps: 200
|
| 50 |
+
```
|