Downloads · 30 days
23
9% of all-time downloads
konstantindobler/mistral7b-de-tokenizer-swap-pure-bf16-v2-anneal-ablation
mistral7b-de-tokenizer-swap-pure-bf16-v2-anneal-ablation is a text generation model from konstantindobler. Use it when you need the model to write or continue text. It is set up for transformers. The card lists the license as apache-2.0.
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".
Downloads · 30 days
23
9% of all-time downloads
All-time downloads
247
Public
Parameters
7.2B
29 GB on disk
Likes
0
Public
Click a slice to open those files.
.safetensors29 GB · 100%
From the Hugging Face model README
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
from transformers import AutoTokenizer, AutoModelForCausalLM
tokenizer = AutoTokenizer.from_pretrained("konstantindobler/mistral7b-de-tokenizer-swap-pure-bf16-v2-anneal-ablation")
model = AutoModelForCausalLM.from_pretrained("konstantindobler/mistral7b-de-tokenizer-swap-pure-bf16-v2-anneal-ablation")
# 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 uonlp/CulturaX with pure bfloat16 precision (no mixed precision). However, in the final annealing phase of the learning rate schedule, the model was again trained using bfloat16 mixed precision. More details and hyperparameters can be found in the paper.
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.
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}
}
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.