Igor Carron

@igorcarron.bsky.social

LightOn, President GenAI for enterprise, LLMs Former Nuclear and Rocket Scientist Nuit Blanche writer (Compressive Sensing, Matrix Factorizations to DL). Paris Machine Learning meetup cofounder

✊ Transformers... Assemble! Introducing ♊ Ettin Suite, a SoTA open recipe to outperform existing Generative & Retrieval Models. Developed by Johns Hopkins University in collaboration with LightOn, Ettin is the first-ever SoTA suite of paired encoder & decoder models.

Bild

✨ Your know-how is your edge, AI is made to enhance it on your terms! 👀 Would you let AI access data you wouldn’t share with your team? 🛡️Internal governance remains paramount! This isn’t just smart permissions, it’s safer intelligence — activated

🎭 Your know-how comes in many forms, scattered across countless formats. 💡 What if the insight you seek - hidden in PDFs, PPTXs, XLSXs, PNGs, and JPEGs - was just one question away? This isn’t just multimodal search, it’s enterprise intelligence — activated

Trying to make sense of visual cues in your company documents so that you can have a richer conversation with them? Introducing visual RAG un Paradigm so that you can make the best of your documents 🚀 in a secure and private way. At @lightonai.bsky.social we're shipping fast and often 🚢

a page showing visual RAG in LightOn paradigm platform. It shows the ability to make a request in words and obtain an answer from a graph.

✨ Introducing the first ever Visual RAG pipeline! Paradigm now offers a whole pipeline working directly on your visual documents as you see them, preventing any loss of information and allowing you to query non-textual elements!

Bild

LightOn releases ModernBERT-embed-large model, a larger and more proficient iteration of ModernBERT-embed-base! This competitive fine-tuned variant aims at allowing everyone to play with ModernBERT, becoming an out-of-the-box tool and a starting point for fine-tuning! huggingface.co/lightonai/mo...

lightonai/modernbert-embed-large · Hugging Face

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

huggingface.co

I am thrilled to announce the release of ModernBERT, the long-awaited BERT replacement! There might be a few LLM releases per week, but there is only one drop-in replacement that brings Pareto improvements over the 6 years old BERT while going at lightspeed

Bild

The FineWeb-2 study evaluated the quality of various pre-training datasets for the Arabic language. It's great to see that our work on ArabicWeb is competitive with the state of the art.

Bild