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".
1from transformers import AutoTokenizer, AutoModelForCausalLM
2
3tokenizer = AutoTokenizer.from_pretrained("konstantindobler/mistral7b-de-tokenizer-swap-pure-bf16")
4model = AutoModelForCausalLM.from_pretrained("konstantindobler/mistral7b-de-tokenizer-swap-pure-bf16")
5
6# Use model and tokenizer as usual
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
oscar-corpus/OSCAR-2301 with pure bfloat16 precision (no mixed precision). More details and hyperparameters can be found
in the paper.
The web-scale dataset used for pretraining and tokenizer training (
oscar-corpus/OSCAR-2301) might contain personal and sensitive information.
Such behavior needs to be assessed carefully before any real-world deployment of the models.
1@inproceedings{dobler2024language,
2 title={Language Adaptation on a Tight Academic Compute Budget: Tokenizer Swapping Works and Pure bfloat16 Is Enough},
3 author={Konstantin Dobler and Gerard de Melo},
4 booktitle={2nd Workshop on Advancing Neural Network Training: Computational Efficiency, Scalability, and Resource Optimization (WANT@ICML 2024)},
5 year={2024},
6 url={https://openreview.net/forum?id=VYfJaHeVod}
7}
The project on which this model is based was funded by the Federal Ministry of Education and Research under the funding code "KI-Servicezentrum Berlin-Brandenburg" 01IS22092. Responsibility for the content of this publication remains with the author.
