louisbrulenaudet/lemone-router-s

Lemone-Router: A Series of Fine-Tuned Classification Models for French Taxation

Lemone-router is a series of classification models designed to produce an optimal multi-agent system for different branches of tax law. Trained on a base of 49k lines comprising a set of synthetic questions generated by GPT-4 Turbo and Llama 3.1 70B, which have been further refined through evol-instruction tuning and manual curation and authority documents, these models are based on an 8-category decomposition of the classification scheme derived from the Bulletin officiel des finances publiques - impôts :

label2id = {
    "Bénéfices professionnels": 0,
    "Contrôle et contentieux": 1,
    "Dispositifs transversaux": 2,
    "Fiscalité des entreprises": 3,
    "Patrimoine et enregistrement": 4,
    "Revenus particuliers": 5,
    "Revenus patrimoniaux": 6,
    "Taxes sur la consommation": 7
}
    
id2label = {
    0: "Bénéfices professionnels",
    1: "Contrôle et contentieux",
    2: "Dispositifs transversaux",
    3: "Fiscalité des entreprises",
    4: "Patrimoine et enregistrement",
    5: "Revenus particuliers",
    6: "Revenus patrimoniaux",
    7: "Taxes sur la consommation"
}

This model is a fine-tuned version of intfloat/multilingual-e5-small. It achieves the following results on the evaluation set:

• Loss: 0.4446
• Accuracy: 0.9061

Usage

# Load model directly
from transformers import AutoTokenizer, AutoModelForSequenceClassification

tokenizer = AutoTokenizer.from_pretrained("louisbrulenaudet/lemone-router-s")
model = AutoModelForSequenceClassification.from_pretrained("louisbrulenaudet/lemone-router-s")

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 9.736038436571022e-05
• train_batch_size: 8
• eval_batch_size: 64
• seed: 6
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 4

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
0.6002	1.0	5617	0.5964	0.8207
0.4506	2.0	11234	0.4375	0.8788
0.3138	3.0	16851	0.4400	0.8965
0.189	4.0	22468	0.4446	0.9061

Training Hardware

• On Cloud: No
• GPU Model: 1 x NVIDIA H100 NVL
• CPU Model: AMD EPYC 9V84 96-Core Processor

Framework versions

• Transformers 4.45.2
• Pytorch 2.4.1+cu121
• Datasets 2.21.0
• Tokenizers 0.20.1

Citation

If you use this code in your research, please use the following BibTeX entry.

@misc{louisbrulenaudet2024,
  author =       {Louis Brulé Naudet},
  title =        {Lemone-Router: A Series of Fine-Tuned Classification Models for French Taxation},
  year =         {2024}
  howpublished = {\url{https://huggingface.co/datasets/louisbrulenaudet/lemone-router-s}},
}

Feedback

If you have any feedback, please reach out at louisbrulenaudet@icloud.com.