fietje-2-chat

<p align="center" style="margin:0;padding:0"> <img src="https://huggingface.co/BramVanroy/fietje-2-chat/resolve/main/img/fietje-2b-banner-rounded.png" alt="Fietje banner" width="800" style="margin-left:'auto' margin-right:'auto' display:'block'"/> </p> <div style="margin:auto; text-align:center"> <h1 style="margin-bottom: 0">Fietje 2 Chat</h1> <em>An open and efficient LLM for Dutch</em> </div> <blockquote class="tip" style="padding: 1.5em; border: 0"> <p align="center" style="text-align: center; margin: 0"> <a href="https://huggingface.co/BramVanroy/fietje-2">👱‍♀️ Base version</a> - <a href="https://huggingface.co/BramVanroy/fietje-2-instruct">🤖 Instruct version</a> - <a href="https://huggingface.co/BramVanroy/fietje-2-chat">💬 Chat version</a> (this one) - <a href="https://huggingface.co/BramVanroy/fietje-2-chat-GGUF">🚀 GGUF of Chat</a> </p> </blockquote>

This is the chat version of Fietje, a DPO-tuned (aligned) continuation on the instruct version. Fietje is an adapated version of microsoft/phi-2, tailored to Dutch text generation by training on 28B tokens. It is small and efficient with a size of 2.7 billion parameters while performing almost on par with more powerful Dutch LLMs of twice its size like GEITje 7B Ultra.

A thorough description of the creation and evaluation of Fietje as well as usage examples are available in this Github repository.

Citation

If you use Fietje or the CulturaX + Wikipedia filtered subset in your work, please cite to the following paper:

@article{vanroy2024fietje,
    author="Vanroy, Bram",
    title="Fietje: An open, efficient LLM for Dutch",
    url="https://www.clinjournal.org/clinj/article/view/213",
    journal="Computational Linguistics in the Netherlands Journal",
    volume="14",
    year="2025",
    pages="473--504"
}

Intended uses & limitations

The same limitations as phi-2, and LLMs in general, apply here. LLMs hallucinate, make mistakes, and should not be trusted. Use at your own risk!

Training and evaluation data

Fietje 2 Chat was finetuned from the instruct model on the following datasets. Number of training samples per dataset given in brackets, totalling 18,653 samples.

• BramVanroy/ultra_feedback_dutch_cleaned subset dpo_hq: a cleaned version of BramVanroy/ultra_feedback_dutch (9186)
• BramVanroy/orca_dpo_pairs_dutch_cleaned subset dpo_all: a cleaned version of BramVanroy/orca_dpo_pairs_dutch (9467)

A lot of different learning rates, beta, en batch sizes were investigated in search of a converging combination. You can find them all in the W&B runs.

Training procedure

I am thankful to the Flemish Supercomputer Center (VSC) for providing the computational power to accomplish this project. Accounting for waiting for jobs, training a single run took around nine hours on one A100 80GB.

Training was done with the wonderful alignment-handbook, using DeepSpeed as a back-end. Exact training recipes and SLURM script are given in the Github repository.

Training hyperparameters

The following hyperparameters were used during training:

• beta: 0.2
• learning_rate: 2e-06
• train_batch_size: 8
• eval_batch_size: 4
• seed: 42
• distributed_type: multi-GPU
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.98) and epsilon=1e-07
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 1.0

Training results

Framework versions

• Transformers 4.39.1
• Pytorch 2.1.2+cu121
• Datasets 2.18.0
• Tokenizers 0.15.2

License

21/10/2026: the no_robots_dutch dataset was derived from a non-commercially available dataset. Its license was initally incorrectly specified as Apache 2.0, which allowed Fietje to instruct/chat to be released under MIT. However, this has now been rectified: Fietje Instruct and Fietje Chat may not be used for commercial purposes.