atsuki-yamaguchi/Llama-3.1-8B-gu-madlad-mean-cv

Llama 3.1 8B for Gujarati: Chat Vector

This model is built on top of Llama 3.1 8B adapted for Gujarati using 500M target language tokens sampled from MADLAD-400. It has an additional target vocabulary of 10K. Chat vector was added to the model after continual pre-training.

Model Details

• Vocabulary: This model has an additional target vocabulary of 10K.
• Target vocabulary initialization: The target weights of the embedding and LM head were initialized using mean initialization.
• Training: This model was continually pre-trained on 500M target language tokens sampled from MADLAD-400.
• Post-processing: The model was post-processed using the Chat Vector method.

Model Description

• Language: Gujarati
• License: Llama 3.1 Community License Agreement
• Fine-tuned from model: meta-llama/Llama-3.1-8B

Model Sources

• Repository: https://github.com/gucci-j/chat-cve
• Paper: https://arxiv.org/abs/2412.11704

How to Get Started with the Model

Use the code below to get started with the model.

from transformers import AutoTokenizer, AutoModelForCausalLM

model = AutoModelForCausalLM.from_pretrained(
    "atsuki-yamaguchi/Llama-3.1-8B-gu-madlad-mean-cv"
)
tokenizer = AutoTokenizer.from_pretrained(
    "atsuki-yamaguchi/Llama-3.1-8B-gu-madlad-mean-cv"
)

Citation

@misc{yamaguchi2024vocabularyexpansionchatmodels,
      title={{ElChat}: Adapting Chat Language Models Using Only Target Unlabeled Language Data}, 
      author={Atsuki Yamaguchi and Terufumi Morishita and Aline Villavicencio and Nikolaos Aletras},
      year={2024},
      eprint={2412.11704},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2412.11704},
}