# Model Summary

Llama3-70B-Chinese-Chat is **one of the first instruction-tuned LLMs for Chinese & English users with various abilities** such as roleplaying, tool-using, and math, built upon the [meta-llama/Meta-Llama-3-70B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3-70B-Instruct) model.

Developed by: [Shenzhi Wang](https://shenzhi-wang.netlify.app) (王慎执) and [Yaowei Zheng](https://github.com/hiyouga) (郑耀威)

- License: [Llama-3 License](https://llama.meta.com/llama3/license/)
- Base Model: Meta-Llama-3-70B-Instruct
- Model Size: 70.6B
- Context length: 8K

# 1. Introduction

This is **one of the first LLM fine-tuned specifically for Chinese and English users**, based on the [Meta-Llama-3-70B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3-70B-Instruct) model. The fine-tuning algorithm used is **ORPO** [1].

Our Llama3-70B-Chinese-Chat model was trained on a dataset containing **over 100K preference pairs**, with a roughly equal ratio of Chinese and English data. This dataset will be available soon.

**Compared to the original [Meta-Llama-3-70B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3-70B-Instruct) model, the Llama3-70B-Chinese-Chat model greatly reduces the issues of "Chinese questions with English answers" and the mixing of Chinese and English in responses. Additionally, Llama3-70B-Chinese-Chat excels at roleplaying, function calling, and mathematics.**

With much more parameters than our [Llama3-8B-Chinese-Chat](https://huggingface.co/shenzhi-wang/Llama3-8B-Chinese-Chat) model, our Llama3-70B-Chinese-Chat offers significant performance enhancements. If you enjoyed our [Llama3-8B-Chinese-Chat](https://huggingface.co/shenzhi-wang/Llama3-8B-Chinese-Chat), the Llama3-70B-Chinese-Chat is a must-try!

[1] Hong, Jiwoo, Noah Lee, and James Thorne. "Reference-free Monolithic Preference Optimization with Odds Ratio." arXiv preprint arXiv:2403.07691 (2024).

Training framework: [LLaMA-Factory](https://github.com/hiyouga/LLaMA-Factory).

Training details:

- epochs: 3 (We also provide a 2-epoch model version at the [`epoch_2` branch](https://e.gitee.com/wang-shenzhi/repos/wang-shenzhi/llama3-70b-chinese-chat/tree/epoch_2))
- learning rate: 1.5e-6
- learning rate scheduler type: cosine
- Warmup ratio: 0.1
- cutoff len (i.e. context length): 8192
- orpo beta (i.e. $\lambda$ in the ORPO paper): 0.05
- global batch size: 128
- fine-tuning type: full parameters
- optimizer: paged_adamw_32bit

# 2. Usage

```python
from transformers import AutoTokenizer, AutoModelForCausalLM

model_id = "shenzhi-wang/Llama3-70B-Chinese-Chat"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id, torch_dtype="auto", device_map="auto"
)

messages = [
    {"role": "user", "content": "写一首诗吧"},
]

input_ids = tokenizer.apply_chat_template(
    messages, add_generation_prompt=True, return_tensors="pt"
).to(model.device)

outputs = model.generate(
    input_ids,
    max_new_tokens=8192,
    do_sample=True,
    temperature=0.6,
    top_p=0.9,
)
response = outputs[0][input_ids.shape[-1]:]
print(tokenizer.decode(response, skip_special_tokens=True))
```