CallMeDaniel/Llama-2-7b-chat-hf_vn

Vietnamese Fine-tuned Llama-2-7b-chat-hf

This repository contains a Vietnamese-tuned version of the Llama-2-7b-chat-hf model, which has been fine-tuned on Vietnamese datasets using LoRA (Low-Rank Adaptation) techniques.

Model Details

This model is a fine-tuned version of the Llama-2-7b-chat-hf model, specifically adapted for improved performance on Vietnamese language tasks. It uses LoRA fine-tuning to efficiently adapt the large language model to Vietnamese data while maintaining much of the original model's general knowledge and capabilities.

Model Description

• Developed by: Daniel Du
• Model type: Large Language Model
• Language(s) (NLP): Vietnamese
• License: [More Information Needed]
• Finetuned from model [optional]: meta-llama/Llama-2-7b-chat-hf
• Language: Vietnamese

Direct Use

You can use this model directly with the Hugging Face Transformers library:

from transformers import AutoTokenizer, AutoModelForCausalLM
from peft import PeftModel, PeftConfig

# Load the base model
base_model = AutoModelForCausalLM.from_pretrained("meta-llama/Llama-2-7b-chat-hf")

# Load the LoRA configuration and model
peft_model_id = "CallMeMrFern/Llama-2-7b-chat-hf_vn"
config = PeftConfig.from_pretrained(peft_model_id)
model = PeftModel.from_pretrained(base_model, peft_model_id)

# Load the tokenizer
tokenizer = AutoTokenizer.from_pretrained("meta-llama/Llama-2-7b-chat-hf")

# Example usage
input_text = "Xin chào, hôm nay thời tiết thế nào?"
inputs = tokenizer(input_text, return_tensors="pt")
outputs = model.generate(**inputs, max_length=100)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Downstream Use [optional]

[More Information Needed]

Out-of-Scope Use

[More Information Needed]

Bias, Risks, and Limitations

• This model is specifically fine-tuned for Vietnamese and may not perform as well on other languages.
• The model inherits limitations from the base Llama-2-7b-chat-hf model.
• Performance may vary depending on the specific task and domain.

Recommendations

Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.

How to Get Started with the Model

Use the code below to get started with the model.

[More Information Needed]

Training Details

Training Data

Dataset: alpaca_translate_GPT_35_10_20k.json (Vietnamese translation of the Alpaca dataset)

Training Hyperparameters

• Training regime: [More Information Needed]

Speeds, Sizes, Times [optional]

[More Information Needed]

Evaluation

Testing Data, Factors & Metrics

Testing Data

[More Information Needed]

Factors

[More Information Needed]

Metrics

[More Information Needed]

Results

Summary

Model Examination [optional]

[More Information Needed]

Environmental Impact

Carbon emissions can be estimated using the Machine Learning Impact calculator presented in Lacoste et al. (2019).

• Hardware Type: [More Information Needed]
• Hours used: [More Information Needed]
• Cloud Provider: [More Information Needed]
• Compute Region: [More Information Needed]
• Carbon Emitted: [More Information Needed]

Model Architecture and Objective

[More Information Needed]

Citation

If you use this model in your research, please cite:

@misc{vietnamese_llama2_7b_chat,
  author = {[Your Name]},
  title = {Vietnamese Fine-tuned Llama-2-7b-chat-hf},
  year = {2023},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished = {\url{https://huggingface.co./CallMeMrFern/Llama-2-7b-chat-hf_vn}}
}

Training procedure

The following bitsandbytes quantization config was used during training:

• quant_method: bitsandbytes
• load_in_8bit: True
• load_in_4bit: False
• llm_int8_threshold: 6.0
• llm_int8_skip_modules: None
• llm_int8_enable_fp32_cpu_offload: False
• llm_int8_has_fp16_weight: False
• bnb_4bit_quant_type: fp4
• bnb_4bit_use_double_quant: False
• bnb_4bit_compute_dtype: float32

Framework versions

• PEFT 0.6.3.dev0

Model Description

This model is a fine-tuned version of the Llama-2-7b-chat-hf model, specifically adapted for improved performance on Vietnamese language tasks. It uses LoRA fine-tuning to efficiently adapt the large language model to Vietnamese data while maintaining much of the original model's general knowledge and capabilities.

Fine-tuning Details

• Fine-tuning Method: LoRA (Low-Rank Adaptation)
• LoRA Config:
  • Target Modules: ["q_proj", "v_proj"]
  • Precision: 8-bit
• Dataset: alpaca_translate_GPT_35_10_20k.json (Vietnamese translation of the Alpaca dataset)

Training Procedure

The model was fine-tuned using the following command:

python finetune/lora.py \
--base_model meta-llama/Llama-2-7b-chat-hf \
--model_type llama \
--data_dir data/general/alpaca_translate_GPT_35_10_20k.json \
--output_dir finetuned/meta-llama/Llama-2-7b-chat-hf \
--lora_target_modules '["q_proj", "v_proj"]' \
--micro_batch_size 1

For multi-GPU training, a distributed training approach was used.

Evaluation Results

[Include any evaluation results, perplexity scores, or benchmark performances here]

Acknowledgements

• This project is part of the TF07 Course offered by ProtonX.
• We thank the creators of the original Llama-2-7b-chat-hf model and the Hugging Face team for their tools and resources.
• Appreciation to VietnamAIHub/Vietnamese_LLMs for the translated dataset.