README.md

---
library_name: peft
base_model: chihoonlee10/T3Q-Mistral-Orca-Math-DPO
---

# Model Card for Model ID
推理配置需要注意的几个参数：
```
params = {
  'temperature': 0.85,
  'top_p': 0.95,
  'top_k': 20,
  'repetition_penalty': 1.18,
  'max_tokens': 500,
  'stop': [],
  'typical_p': 0.95,
  'n': 1,
}
```

Prompt模板格式
```
### Instruction:

<prompt> (without the <>)

### Response:
```

训练参数(使用Llama-Factory训练)：
```
- learning_rate: 5e-05
- lr_scheduler_type: cosine
- per_device_train_batch_size: 1
- per_device_eval_batch_size: 1
- gradient_accumulation_steps: 2
- warmup_steps: 24
- num_train_epochs: 2
- template: alpaca
- cutoff_len: 4096
- finetuning_type: lora
- lora_target: q_proj,v_proj,o_proj,k_proj
- quantization_bit: 4
- lora_rank: 64
- lora_alpha: 16
- bf16: True
- logging_steps: 20
- val_size: 4
- save_steps: 200
```