mistral7b-de-tokenizer-swap-pure-bf16-v2

Mistral-7B-v0.1 adapted to German as part of our study on efficient language adaptation: "Language Adaptation on a Tight Academic Compute Budget: Tokenizer Swapping Works and Pure bfloat16 Is Enough".

Code: https://github.com/konstantinjdobler/tight-budget-llm-adaptation

Paper: https://openreview.net/forum?id=VYfJaHeVod

Usage

from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("konstantindobler/mistral7b-de-tokenizer-swap-pure-bf16-v2")
model = AutoModelForCausalLM.from_pretrained("konstantindobler/mistral7b-de-tokenizer-swap-pure-bf16-v2")

# Use model and tokenizer as usual

Details

The model is based on Mistral-7B-v0.1 and was adapted to German. The original tokenizer was replaced by a language-specific German tokenizer with a vocabulary of 32768 tokens. The new embeddings were initialized with FOCUS. The model was then trained on 8 billion German tokens from uonlp/CulturaX with pure bfloat16 precision (no mixed precision). More details and hyperparameters can be found in the paper.

Disclaimer

The web-scale dataset used for pretraining and tokenizer training (uonlp/CulturaX) might contain personal and sensitive information. Such behavior needs to be assessed carefully before any real-world deployment of the models.

Citation

Please cite as follows:

@inproceedings{dobler2024language,
    title={Language Adaptation on a Tight Academic Compute Budget: Tokenizer Swapping Works and Pure bfloat16 Is Enough},
    author={Konstantin Dobler and Gerard de Melo},
    booktitle={2nd Workshop on Advancing Neural Network Training: Computational Efficiency, Scalability, and Resource Optimization (WANT@ICML 2024)},
    year={2024},
    url={https://openreview.net/forum?id=VYfJaHeVod}
}