Lemone

non-profit

Verified

https://lemone.io

AI & ML interests

La recherche ouverte sur le traitement automatique du langage, dédiée à la matière fiscale 🔬

Recent Activity

louisbrulenaudet updated a Space 11 months ago

lemoneresearch/README

louisbrulenaudet updated a model about 1 year ago

lemoneresearch/tsdae-lemone-mbert-tax

View all activity

lemoneresearch's activity

louisbrulenaudet

posted an update about 1 month ago

Post

1731

I’ve published a new dataset to simplify model merging 🤗

This dataset facilitates the search for compatible architectures for model merging with @arcee_ai’s mergekit, streamlining the automation of high-performance merge searches 📖

Dataset : louisbrulenaudet/mergekit-configs

1 reply

·

louisbrulenaudet

posted an update about 2 months ago

Post

1141

Introducing Lemone-router, 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"
}

It achieves the following results on the evaluation set:
- Loss: 0.4734
- Accuracy: 0.9191

Link to the collection: louisbrulenaudet/lemone-router-671cce21d6410f3570514762

louisbrulenaudet

posted an update 2 months ago

Post

3105

🚨 I have $3,500 in Azure credits, including access to an H100 (96 Go), expiring on November 12, 2024.

I won’t be able to use it all myself, so I’m reaching out to the @huggingface community: Are there any open-source projets with data ready for some compute power?

Let’s collaborate and make the most of it together 🔗

9 replies

·

louisbrulenaudet

posted an update 3 months ago

Post

2099

My biggest release of the year: a series of 7 specialized embedding models for information retrieval within tax documents, is now available for free on Hugging Face 🤗

These new models aim to offer an open source alternative for in-domain semantic search from large text corpora and will improve RAG systems and context addition for large language models.

Trained on more than 43 million tax tokens derived from semi-synthetic and raw-synthetic data, enriched by various methods (in particular MSFT's evol-instruct by @intfloat ), and corrected by humans, this project is the fruit of hundreds of hours of work and is the culmination of a global effort to open up legal technologies that has only just begun.

A big thank you to Microsoft for Startups for giving me access to state-of-the-art infrastructure to train these models, and to @julien-c , @clem 🤗, @thomwolf and the whole HF team for the inference endpoint API and the generous provision of Meta LLama-3.1-70B. Special thanks also to @tomaarsen for his invaluable advice on training embedding models and Loss functions ❤️

Models are available on my personal HF page, into the Lemone-embed collection: louisbrulenaudet/lemone-embed-66fdc24000df732b395df29b

1 reply

·

louisbrulenaudet

posted an update 4 months ago

Post

2595

The Romulus model series has been released on Hugging Face, continually pre-trained on 34,864,949 tokens of French laws and intended to serve as a foundation for fine-tuning on labeled data 🤗

The training code, dataset and model weights are open and available free on HF and the training was based on H100 provided by Microsoft for Startups using Unsloth AI by @danielhanchen and @shimmyshimmer 🦥

Link to the base model: louisbrulenaudet/Romulus-cpt-Llama-3.1-8B-v0.1

Link to the instruct model: louisbrulenaudet/Romulus-cpt-Llama-3.1-8B-v0.1-Instruct

Link to the dataset: louisbrulenaudet/Romulus-cpt-fr

Please note that these models have not been aligned for the production of usable texts as they stand, and will certainly need to be refined for the desired tasks in order to produce satisfactory results.

1 reply

·

louisbrulenaudet

posted an update 4 months ago

Post

1575

An example of the application of LegalKit is the production of knowledge graphs, here is a demo Space 🔗

With the update of the French legal code data model uploaded to 🤗 and the introduction of a column dedicated to HTML text, it's now easy to extract links between different articles and produce complex graphs with just a few lines of Python.

This simplified demo highlights the ease of implementation and creative potential, and enables the generation of complete data sets, although requiring a powerful graphics card for display. The framework used for the moment is D3.js, but perhaps other solutions are possible. I'd be delighted to hear your suggestions, and look forward to hearing from the community.

Link to the 🤗 Space: louisbrulenaudet/legalkit-knowledge-graph

2 replies

·

louisbrulenaudet

posted an update 4 months ago

Post

1937

Understanding the json format response with HF's Serverless Inference API 🤗

As it stands, there seems to be an inconsistency with the OpenAI documentation on the question of implementing the JSON response format using the InferenceClient completion API.

After investigating the InferenceClient source code, I share the official solution using a JSON Schema. This consolidates the structure of the response and simplifies parsing as part of an automated process for extracting metadata, information:

from huggingface_hub import InferenceClient

client = InferenceClient("meta-llama/Meta-Llama-3-70B-Instruct")

messages = [
    {
        "role": "user",
        "content": "I saw a puppy a cat and a raccoon during my bike ride in the park. What did I saw and when?",
    },
]

response_format = {
    "type": "json",
    "value": {
        "properties": {
            "location": {"type": "string"},
            "activity": {"type": "string"},
            "animals_seen": {"type": "integer", "minimum": 1, "maximum": 5},
            "animals": {"type": "array", "items": {"type": "string"}},
        },
        "required": ["location", "activity", "animals_seen", "animals"],
    },
}

response = client.chat_completion(
    messages=messages,
    response_format=response_format,
    max_tokens=500,
)

print(response.choices[0].message.content)

As a reminder, json mode is activated with the OpenAI client as follows:

response = client.chat.completions.create(
     model="gpt-3.5-turbo-0125",
     messages=[...],
     response_format={"type": "json_object"}
)

One question remains unanswered, however, and will perhaps be answered by the community: it seems that an incompatibility persists for list of dictionaries generation, and currently, the production of simple dictionaries seems to be the only functional option.

2 replies

·

louisbrulenaudet

posted an update 5 months ago

Post

2760

🚀 RAGoon is now available on PyPI, GitHub, and as a Space on Hugging Face for batched embeddings generation 🤗

RAGoon is a set of NLP utilities for multi-model embedding production, high-dimensional vector visualization, and aims to improve language model performance by providing contextually relevant information through search-based querying, web scraping and data augmentation techniques.

At this stage, 5 major classes are available via RAGoon to facilitate:
- the production of chain embeddings for several models to simplify a continuous deployment process;
- production of LLM requests for web querying and content retrieval via the Google API;
- recursive chunking via tokens;
- data visualization and the function to load embeddings from a FAISS index, reduce their dimensionality using PCA and/or t-SNE, and visualize them in an interactive 3D graph;
- the creation of binary indexes for search with scalar (int8) rescoring.

Link to GitHub: https://github.com/louisbrulenaudet/ragoon
Link to the 🤗 Space: louisbrulenaudet/ragoon

louisbrulenaudet

posted an update 5 months ago

Post

869

You can now find the OBIS - Ocean Biodiversity Information System, on Hugging Face with 128M rows, via the Datasets package stream 🤗

The datasets are integrated, allowing seamless search and mapping by species name, higher taxonomic level, geographic area, depth, time, and environmental parameters. OBIS originates from the Census of Marine Life (2000-2010) and was adopted as a project under IOC-UNESCO’s International Oceanographic Data and Information (IODE) programme in 2009.

Collectively, they have provided over 45 million observations of nearly 120,000 marine species, ranging from bacteria to whales, from the surface to 10,900 meters depth, and from the tropics to the poles.

Link to the dataset: louisbrulenaudet/obis

louisbrulenaudet

posted an update 6 months ago

Post

2116

Introducing the first two projects on the HFforLegal community: the 'Laws' dataset and the associated search tool based on @nreimers and @tomaarsen 's Sentence Transformers library 🤗

The objective of these two tools is to centralize in a single format a set of rules from different countries and legal systems in order to facilitate NLP in the field of comparative law, enabling more accurate and comprehensive legal analysis across different jurisdictions 🌍

Link to the dataset : HFforLegal/laws
Link to the space: HFforLegal/laws-retrieval

We need your contributions to enrich this new knowledge base, and you will find in the 'Laws' dataset all the information you need to format your data and submit them to the appropriate split.

louisbrulenaudet

posted an update 6 months ago

Post

2947

Announcing the creation of the "HF for Legal" organization, an open-source community dedicated to demystifying language models for legal professionals 🤗

Whether you're a practicing attorney, a legal scholar, or a technologist interested in legal applications of AI, HF for Legal may be your hub for exploration, learning, and free innovation ⚗️

On the occasion of this launch, you'll be able to find several notebooks I've been developing over the last few months for TSDAE pre-training of embedding models, the generation of indexes for semantic search, based on the formidable work of @tomaarsen and @nreimers , adapted to the field of French law, or the addition of information retrieval tasks to the MTEB.

Join us in our mission to make AI more accessible and understandable for the legal world, ensuring that the power of language models can be harnessed effectively and ethically.

Link to the org: https://huggingface.co./HFforLegal

Special thanks to @clem for encouraging me to start this organization. Let's hope we can bring together all the enthusiasts who work in this field.

Let's code and share together! 🚀🔗

louisbrulenaudet

posted an update 6 months ago

Post

3233

I am delighted to announce the publication of my LegalKit, a French labeled dataset built for legal ML training 🤗

This dataset comprises multiple query-document pairs (+50k) curated for training sentence embedding models within the domain of French law.

The labeling process follows a systematic approach to ensure consistency and relevance:
- Initial Query Generation: Three instances of the LLaMA-3-70B model independently generate three different queries based on the same document.
- Selection of Optimal Query: A fourth instance of the LLaMA-3-70B model, using a dedicated selection prompt, evaluates the generated queries and selects the most suitable one.
- Final Label Assignment: The chosen query is used to label the document, aiming to ensure that the label accurately reflects the content and context of the original text.

Dataset: louisbrulenaudet/legalkit

Stay tuned for further updates and release information 🔥

@clem , if we can create an "HF for Legal" organization, similar to what exists for journalists, I am available!

Note : My special thanks to @alvdansen for their illustration models ❤️

2 replies

·

louisbrulenaudet

posted an update 6 months ago

Post

4059

Mixtral or Llama 70B on Google Spreadsheet thanks to Hugging Face's Serverless Inference API 🤗

The Add-on is now available on the HF repo "Journalists on Hugging Face" and allows rapid generation of synthetic data, automatic translation, answering questions and more from simple spreadsheet cells 🖥️

Link to the 🤗 Space : JournalistsonHF/huggingface-on-sheets

Although this tool was initially developed for journalists, it actually finds a much wider inking among daily users of the Google suite and the remaining use cases to be explored are numerous.

Only a free Hugging Face API key is required to start using this no-code extension.

Do not hesitate to submit ideas for features that we could add!

Thanks to @fdaudens for initiating this development.

4 replies

·

louisbrulenaudet

posted an update 7 months ago

Post

981

I've just open sourced RAGoon, a small utility I use to integrate knowledge from the web into LLM inference based on Groq speed and pure Google search performance ⚡

RAGoon is a Python library available on PyPI that aims to improve the performance of language models by providing contextually relevant information through retrieval-based querying, parallel web scraping, and data augmentation techniques. It offers an integration of various APIs (OpenAI, Groq), enabling users to retrieve information from the web, enrich it with domain-specific knowledge, and feed it to language models for more informed responses.

from groq import Groq
# from openai import OpenAI
from ragoon import RAGoon

# Initialize RAGoon instance
ragoon = RAGoon(
    google_api_key="your_google_api_key",
    google_cx="your_google_cx",
    completion_client=Groq(api_key="your_groq_api_key")
)

# Search and get results
query = "I want to do a left join in python polars"
results = ragoon.search(
    query=query,
    completion_model="Llama3-70b-8192",
)

# Print list of results
print(results)

For the time being, this project remains simple, but can easily be integrated into a RAG pipeline.

Link to GitHub : https://github.com/louisbrulenaudet/ragoon

louisbrulenaudet

posted an update 7 months ago

Post

534

Integrating the French Taxation Embedding Benchmark Task (beta) into the MTEB 🤗

I'm excited to announce an integration of the French Taxation Embedding Benchmark task into the Massive Text Embedding Benchmark (MTEB).

This addition expands the diverse set of tasks available within MTEB, enabling researchers and practitioners to develop and evaluate retrieval models focused on retrieving relevant tax articles or content based on provided queries.

Link to the 🤗 Dataset : louisbrulenaudet/tax-retrieval-benchmark

Link to the GitHub repo : https://github.com/louisbrulenaudet/tax-retrieval-benchmark

Notes:
The Massive Text Embedding Benchmark for French Taxation and the Dataset are currently in beta and may not be suitable for direct use in production. The size of the Dataset may not be sufficient to handle a wide range of queries and scenarios encountered in real-world settings.

As the Dataset grows and matures, I will provide updates and guidance on its suitability for production use cases.

louisbrulenaudet

posted an update 9 months ago

Post

2258

LegalKit Retrieval, a binary Search with Scalar (int8) Rescoring through French legal codes is now available as a 🤗 Space.

This process is designed to be memory efficient and fast, with the binary index being small enough to fit in memory and the int8 index being loaded as a view. Additionally, the binary index is much faster (up to 32x) to search than the float32 index, while the rescoring is also extremely efficient.

This space also showcases the tsdae-lemone-mbert-base, a sentence embedding model based on BERT fitted using Transformer-based Sequential Denoising Auto-Encoder for unsupervised sentence embedding learning with one objective : french legal domain adaptation.

Link to the 🤗 Space : louisbrulenaudet/legalkit-retrieval

Notes:
The SentenceTransformer model currently in use is in beta and may not be suitable for direct use in production.

2 replies

·

louisbrulenaudet

posted an update 9 months ago

Post

2317

To date, louisbrulenaudet/Maxine-34B-stock is the "Best 🤝 base merges and moerges model of around 30B" on the Open LLM Leaderboard ❤️‍🔥

It is a practical application of the stock method recently implemented by @arcee-ai in the MergeKit :

models:
    - model: ConvexAI/Luminex-34B-v0.2
    - model: fblgit/UNA-34BeagleSimpleMath-32K-v1
merge_method: model_stock
base_model: abacusai/Smaug-34B-v0.1
dtype: bfloat16

Model : louisbrulenaudet/Maxine-34B-stock
LLM Leaderboard best models ❤️‍🔥 Collection : https://huggingface.co./collections/open-llm-leaderboard/llm-leaderboard-best-models-652d6c7965a4619fb5c27a03

louisbrulenaudet

updated a Space 11 months ago

README

louisbrulenaudet

updated a model about 1 year ago

lemoneresearch/tsdae-lemone-mbert-tax

Sentence Similarity • Updated Dec 24, 2023 • 14