atsuki-yamaguchi/gemma-2-9b-te-30K-vr-focus

Gemma2 9B for Telugu: 32K vocabulary replacement + FOCUS target vocabulary initialization + 2x2LS/MTP/512 training

This model is built on top of Gemma2 9B, adapted for Telugu using 30K target language sentences sampled from CC-100.

Model Details

• —Vocabulary: This model has a 32K vocabulary trained on Telugu 30K sentences.
• —Target vocabulary initialisation: The target weights of the embedding were initialised using Mean initialisation.
• —Training: This model was additionally pre-trained on 30K target language sentences sampled from CC-100. The training was conducted with the 2x2LS/MTP/512 strategies introduced in the paper.

Model Description

• —Language: Telugu
• —License: Gemma Terms of Use
• —Fine-tuned from model: google/gemma-2-9b

Model Sources

• —Repository: https://github.com/gucci-j/lowres-cve
• —Paper: https://arxiv.org/abs/2406.11477

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/gemma-2-9b-te-30K-vr-focus"
)
tokenizer = AutoTokenizer.from_pretrained(
    "atsuki-yamaguchi/gemma-2-9b-te-30K-vr-focus"
)

Citation

@article{yamaguchi-etal-2024-effectively,
    title={How Can We Effectively Expand the Vocabulary of LLMs with 0.01GB of Target Language Text?}, 
    author={Atsuki Yamaguchi and Aline Villavicencio and Nikolaos Aletras},
    year={2024},
    journal={ArXiv},
    year={2024},
    volume={abs/2406.11477},
    url={https://arxiv.org/abs/2406.11477}, 
}