FinGEITje-7B-sft

<p align="center" style="margin:0;padding:0"> <img src="https://huggingface.co/snoels/FinGEITje-7B-sft/resolve/main/fingeitje-banner.png" alt="FinGEITje Banner" width="1000"/> </p> <div style="margin:auto; text-align:center"> <h1 style="margin-bottom: 0; font-size: 2em;">🐐 FinGEITje 7B</h1> <em style="font-size: 1em;">A large open Dutch Financial language model.</em> </div>

This model is a fine-tuned version of BramVanroy/GEITje-7B-ultra on the snoels/FinGEITje-sft dataset.

📖 Model Description

FinGEITje 7B is a large open Dutch financial language model with 7 billion parameters, based on Mistral 7B. It has been further trained on Dutch financial texts, enhancing its proficiency in the Dutch language and its knowledge of financial topics. As a result, FinGEITje provides more accurate and relevant responses in the domain of finance.

📊 Training and Evaluation Data

Training Data

FinGEITje 7B was fine-tuned on the snoels/FinGEITje-sft dataset, which consists of translated and processed Dutch financial texts. This dataset includes a wide range of financial topics and instruction tuning data.

Data Processing Steps

1. Translation: Original instruction tuning datasets were translated into Dutch using a specialized translation service to maintain the integrity of financial terminology.
2. Post-processing: The translated data underwent post-processing to correct any translation inconsistencies and to format it according to the original dataset structure.
3. Formatting: The data was formatted to match the style and requirements of instruction tuning datasets, ensuring compatibility with the fine-tuning process.
4. Filtering: A Dutch language check and predefined validation checks were applied to filter out any low-quality or irrelevant data.

Evaluation Data

The model was evaluated using:

• snoels/FinDutchBench: A Dutch financial benchmark dataset designed to assess the model's performance on various financial tasks.

⚙️ Training Procedure

FinGEITje was trained following the methodology described in the Alignment Handbook.

Training Configuration

• The training configuration is based on the recipe outlined in the alignment handbook and can be found in the config_qlora.yaml file.
• The model was further trained using QLoRA (Quantized LoRA) for efficient fine-tuning with reduced computational resources.

Training Hyperparameters

The following hyperparameters were used during training:

• Learning Rate: 0.0002
• Train Batch Size: 4
• Evaluation Batch Size: 8
• Seed: 42
• Distributed Type: Multi-GPU
• Gradient Accumulation Steps: 2
• Total Train Batch Size: 8
• Optimizer: Adam with betas=(0.9, 0.999) and epsilon=1e-08
• LR Scheduler Type: Cosine
• Warmup Ratio: 0.1
• Number of Epochs: 1

Training Results

Evaluation Package

The evaluation package includes a set of metrics defined per task, grouped per dataset to evaluate the model's performance across different financial domains. The evaluation notebooks are available:

• Evaluation in Dutch: Assesses the model's performance on the Dutch financial benchmark dataset.
• Evaluation in English: Evaluates the model's performance on English financial benchmarks for comparison purposes.

Framework Versions

• PEFT: 0.7.1
• Transformers: 4.39.0.dev0
• PyTorch: 2.1.2
• Datasets: 2.14.6
• Tokenizers: 0.15.2

🛠️ How to Use

FinGEITje 7B can be utilized using the Hugging Face Transformers library along with PEFT to load the LoRA adapters efficiently.

Installation

Ensure you have the necessary libraries installed:

pip install torch transformers peft accelerate

Loading the Model

from transformers import AutoTokenizer, AutoModelForCausalLM
from peft import PeftModel

# Load the tokenizer
tokenizer = AutoTokenizer.from_pretrained("BramVanroy/GEITje-7B-ultra", use_fast=False)

# Load the base model
base_model = AutoModelForCausalLM.from_pretrained("BramVanroy/GEITje-7B-ultra", device_map='auto')

# Load the FinGEITje model with PEFT adapters
model = PeftModel.from_pretrained(base_model, "snoels/FinGEITje-7B-sft", device_map='auto')

Generating Text

# Prepare the input
input_text = "Wat zijn de laatste trends in de Nederlandse banksector?"
input_ids = tokenizer.encode(input_text, return_tensors='pt').to(model.device)

# Generate a response
outputs = model.generate(input_ids, max_length=200, num_return_sequences=1)
response = tokenizer.decode(outputs[0], skip_special_tokens=True)

print(response)

🚧 Limitations and Future Work

While FinGEITje 7B demonstrates significant improvements in understanding and generating Dutch financial content, certain limitations exist:

• Data Cutoff: The model's knowledge is limited to the data it was trained on and may not include the most recent developments in the financial sector.
• Accuracy Concerns: The model may generate incorrect or outdated information. Users should verify critical information with reliable sources.
• Biases: Potential biases in the training data may affect the neutrality and fairness of the model's responses.
• Language Scope: Primarily designed for Dutch; performance in other languages is not optimized.
• Ethical Use: Users should ensure that the model's outputs comply with ethical standards and do not promote misinformation or harmful content.

Future Work

• Data Updates: Incorporate more recent and diverse financial datasets to keep the model up-to-date.
• Bias Mitigation: Implement techniques to identify and reduce biases in the model's outputs.
• Performance Enhancement: Fine-tune on more specialized financial topics and complex financial tasks.
• Multilingual Expansion: Extend support to other languages relevant to the financial sector in the Netherlands and Europe.

🙏 Acknowledgements

We would like to thank:

• Rijgersberg (GitHub) for creating GEITje, one of the first Dutch foundation models, and for contributing significantly to the development of Dutch language models.
• Bram Vanroy (GitHub) for creating GEITje-7B-ultra, an open-source Dutch chat model, and for sharing training, translation, and evaluation resources.
• Contributors of the Alignment Handbook for providing valuable resources that guided the development and training process of FinGEITje.
• Silverfin for their collaboration in this research. Silverfin, a Belgian scale-up focused on building an accountancy cloud service, provided valuable insights and resources that were instrumental in the development of FinGEITje. More about their work can be found at Silverfin.

📝 Citation

Link to the paper Link to the arXiv

If you use FinGEITje in your work, please cite:

@inproceedings{10.1145/3677052.3698628,
author = {Noels, Sander and De Blaere, Jorne and De Bie, Tijl},
title = {A Dutch Financial Large Language Model},
year = {2024},
isbn = {9798400710810},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
url = {https://doi.org/10.1145/3677052.3698628},
doi = {10.1145/3677052.3698628},
abstract = {This paper presents FinGEITje, the first Dutch financial Large Language Model (LLM) specifically designed and optimized for various financial tasks. Together with the model, we release a specialized Dutch financial instruction tuning dataset with over 140,000 samples, constructed employing an automated translation and data processing method. The open-source data construction method is provided, facilitating the creation of financial instruction datasets in different languages. To evaluate model performance, the study introduces the first Dutch financial evaluation benchmark, along with an automated evaluation method that utilizes an LLM as an independent evaluator, reducing manual intervention in performance evaluation. The experimental results highlight the superior performance of FinGEITje across five critical Dutch and English financial tasks.},
booktitle = {Proceedings of the 5th ACM International Conference on AI in Finance},
pages = {283–291},
numpages = {9},
keywords = {Financial Large Language Model, Instruction Tuning., Natural Language Processing},
location = {Brooklyn, NY, USA},
series = {ICAIF '24}
}

📜 License

This model is licensed under the Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0) license.

📧 Contact

For any inquiries or questions, please contact Sander Noels.