README.md

---
license: other
datasets:
- vietgpt/wikipedia_vi
- vietgpt/binhvq_news_vi
language:
- vi
- en
pipeline_tag: text-generation
extra_gated_prompt: >-
  Please read the [URA-LLaMA License
  Agreement](https://github.com/martinakaduc/ura-llama-public/blob/main/URA-LLaMa%20Model%20User%20Agreement.pdf)
  before accepting it.
extra_gated_fields:
  Name: text
  Email: text
  Affiliation: text
  Country: text
  I accept the URA-LLaMA License Agreement: checkbox
---
# URA-LLaMa 7B

## Model Details

### Model Description

With a strong commitment to enhancing the quality of large language models for the Vietnamese language, a collaborative effort was undertaken by Vietnamese researchers hailing from Ho Chi Minh University of Technology (HCMUT) - Vietnam National University HCMC and Stanford University. Our endeavor involved the meticulous fine-tuning of Meta LLaMa-2 models using Vietnamese articles sourced from Wikipedia and online news websites. In line with our dedication to fostering community progress, we are pleased to offer our models free of charge for research purposes. For those who wish to delve further into our research and its details, we encourage you to explore the comprehensive information provided below.

- **Developed by:**
  - Duc Q. Nguyen
  - Sang T. Truong
  - Toan D. V. Nguyen
  - Dong D. Le
  - Nhi N. Truong
  - Tho Quan
  - Sanmi Koyejo
- **Funded by:**
  - Microsoft Accelerating Foundation Models Research program
  - Stanford University
  - Ho Chi Minh University of Technology (HCMUT) - VNU-HCM
  - DSciLab (Faculty of Computer Science & Engineering, HCMUT - VNU-HCM)
- **Model type:** Text generation
- **Languages:** Vietnamese, English
- **License:**
  - Custom license available at [LICENSE](https://github.com/martinakaduc/ura-llama-public/blob/main/URA-LLaMa%20Model%20User%20Agreement.pdf)
- **Finetuned from model:** Meta LLaMa-2 70B

### Model Sources

We publicly provide starter source code and access to playground of URA-LLaMa 7B.

- **Repository:** [URA-LLaMa Github](https://github.com/martinakaduc/ura-llama-public)
- **Framework:** [ViLLM](https://github.com/stair-lab/villm)
- **Paper:** Our paper was accepted at NAACL 2024. [Link](https://arxiv.org/abs/2403.02715)

## Uses

This model is primarily designed for text generation. However, as language models, it is versatile and can also function as an encoder for various downstream tasks, akin to other models. For a detailed understanding of its use cases, please refer to the information provided below.

### Direct Use

You can use our models to perform various tasks containing

* Question answering (with context)
* Summarization
* Language modelling
* Text classification
* Translation

### Downstream Use

This model can serve as an encoder for a wide range of downstream tasks, spanning from pure natural language processing to combinations of natural language processing with computer vision or speech processing.

### Out-of-Scope Use

While our models have undergone fine-tuning using extensive Vietnamese datasets, they may not perform optimally in specialized domains necessitating profound domain expertise, such as medicine, politics, chemistry, etc. We kindly request that you refrain from employing our models for political purposes or any endeavors that may cause harm to individuals or compromise the sovereignty and territorial integrity of Vietnam.

## Bias, Risks, and Limitations

Unless required by applicable law, the URA-LLaMa materials and any output and results therefrom are provided on an "as is" basis, without warranties of any kind, either express or implied, including, without limitation, any warranties of title, non-infringement, merchantability, or fitness for a particular purpose. you are solely responsible for determining the appropriateness of using or redistributing the URA-LLaMa materials and assume any risks associated with your use of the URA-LLaMa materials and any output and results.

### Recommendations

Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. In order for the model to work well, you may need perform prompt engineering to create appropriate prompts before inference.

## How to Get Started with the Model

Use the code below to get started with the model.

```python
import transformers
from transformers import AutoModelForCausalLM, AutoTokenizer

pipeline_kwargs={
	"temperature": 1.0,
	"max_new_tokens": 250,
	"top_k": 1,
	"repetition_penalty": 1.1
}
  
if __name__ == "__main__":
    # Load model
    model = AutoModelForCausalLM.from_pretrained(
        "ura-hcmut/ura-llama-7b",
        device_map="auto"
    )
    model.config.pretraining_tp = 1
    model.eval()

    # Load tokenizer
    tokenizer = AutoTokenizer.from_pretrained(
        "ura-hcmut/ura-llama-7b", 
        trust_remote_code=True
    )
    tokenizer.pad_token = tokenizer.eos_token
  
    pipeline = transformers.pipeline(
        model=model, 
        tokenizer=tokenizer,
        return_full_text=False,
        task='text-generation',
        **pipeline_kwargs
    )
  
    query_template = "[INST] <<SYS>>\nBạn là một trợ lý thông minh.\n<</SYS>>\n\nHãy trả lời câu hỏi sau.\nCâu hỏi: {query}\nTrả lời: [/INST]"
  
    while True:
        query = input("Query: ")
        if query == "exit":
            break
      
        query = query_template.format(query=query)
        answer = pipeline(query)[0]["generated_text"]
        print(answer)
```

## Finetuning Details

### Finetuning Data

List of datasets used for finetuning:

* [Vietnamese Wikipedia](https://huggingface.co/datasets/vietgpt/wikipedia_vi)
* [Binhvq News Corpus](https://huggingface.co/datasets/vietgpt/binhvq_news_vi)

### Finetuning Procedure

We utilize the causal language modelling (next token prediction) procedure to finetune our models. Available tutorial is available at [https://huggingface.co/docs/transformers/tasks/language_modeling](https://huggingface.co/docs/transformers/tasks/language_modeling).

#### Finetuning Hyperparameters

- **Training regime:** BFloat16 mixed precision
- **Quantization:** Normal Float 4bit
- **Lora rank:** 128
- **Batch size:** 120
- **Optimizer:** Paged AdamW 32bit
- **Learning rate:** 1e-5

## Evaluation

Our models are tested with various tasks. The detail of evaluation process can be found at our [Leaderboard](https://ai.stanford.edu/~sttruong/villm).


## Environmental Impact

Carbon emissions are estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).

- **Hardware Type:** 1 x A100 80GB
- **Hours used:** ~520h
- **Carbon Emitted:** ~90 kg CO2 eq.

## Citation

If you use URA-LLaMa materials in your research, please cite our  model(s) as below.

**BibTeX:**

```plaintext
@inproceedings{crossing2024,
    title = "Crossing Linguistic Horizons: Finetuning and Comprehensive Evaluation of Vietnamese Large Language Models",
    author = "Truong, Sang T.  and Nguyen, Duc Q.  and Nguyen, Toan D. V.  and Le, Dong D.  and Truong, Nhi N.  and Quan, Tho  and Koyejo, Sanmi",
    booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies",
    month = June,
    year = "2024",
    address = "Seattle, Washington",
    publisher = "Association for Computational Linguistics",
    url = "",
    pages = "",
}
```

## Model Card Authors

## Contact

* Mr. Duc Q. Nguyen: nqduc@hcmut.edu.vn
* Mr. Sang T. Truong: sttruong@cs.stanford.edu
* Assoc. Prof. Tho Quan: qttho@hcmut.edu.vn