Igor Carron
@IgorCarron
Followers
5K
Following
396K
Media
3K
Statuses
226K
CEO https://t.co/b9fz6WvhTx @LightOnIO Paris Machine Learning Meetup (8200+) @ParisMLGroup https://t.co/jY1eeMkqJE (10M+ pageviews) @NuitBlog Rocket Scientist
Joined April 2008
🚀Super proud of the team @LightOnIO for this new release of Paradigm, the GenAI platform for the Enterprise. Luminous Lynx marks a major leap forward, integrating advanced agentic 🤖 behavior and cutting-edge tools 🔨✂️🪛🪚. Wanna more ⬇️.
✨ Paradigm Luminous Lynx Release is live!. Smarter, sharper and more frugal, Luminous Lynx combines State-of-The-Art Performance with Cost-Effectiveness. Introducing agentic behavior in your Private GenAI. You can now search and analyze your documents or the web in Paradigm,
0
5
13
RT @orionweller: 🤔 Have you ever wondered how good ModernBERT is compared to decoders like Llama?. We made an open-data version of ModernBE….
0
39
0
RT @LightOnIO: ✊ Transformers. Assemble!. Introducing ♊Ettin Suite, a SoTA open recipe to outperform existing Generative & Retrieval Mode….
0
11
0
🚀Transformers. Assemble!
✊ Transformers. Assemble!. Introducing ♊Ettin Suite, a SoTA open recipe to outperform existing Generative & Retrieval Models. Developed by @JohnsHopkins in collaboration with @LightOnIO, Ettin is the first-ever SoTA suite of paired encoder & decoder models. The revolution
0
0
2
Je ne vois qu'une invention française. Et pourtant, nous avons créé des industries entières.
Engineering is the closest thing to magic we have in this world
0
0
0
Hey @Kimi_Moonshot , any intention on releasing k2 at 11T and some data ?. Awesome work by the way.
David,. What is potentially interesting is the ability to further train a checkpoint of @Kimi_Moonshot K2 starting at 11T because of WSD!.
0
0
0
David,. What is potentially interesting is the ability to further train a checkpoint of @Kimi_Moonshot K2 starting at 11T because of WSD!.
Every ML Engineer’s dream loss curve:. “Kimi K2 was pre-trained on 15.5T tokens using MuonClip with zero training spike, demonstrating MuonClip as a robust solution for stable, large-scale LLM training.”.
1
0
14
RT @antoine_chaffin: @thoDdnn @HamelHusain It is on YouTube!. Not sure why it is not on Maven nor if it will be.
0
6
0
Wow!.
Every ML Engineer’s dream loss curve:. “Kimi K2 was pre-trained on 15.5T tokens using MuonClip with zero training spike, demonstrating MuonClip as a robust solution for stable, large-scale LLM training.”.
0
0
1
oui.
Our single most highly concentrated energy source is the thermonuclear bomb, perfected decades ago and still our most potent technological avatar. It is our Shiva, a meditating destroyer-god and concierge to the most creative period in our history. There are just under four
0
0
0
RT @LightOnIO: 🏎️ Introducing PyLate-rs. @raphaelsrty is back at it to make you love late-interaction models!. After quickening the retriev….
0
16
0
RT @LightOnIO: AI is already running in your company, the good news is that you can secure it!. LightOn puts Generative AI where it belongs….
0
5
0
Une deux-cheveaux, vous avez recréé une deux-cheveaux!.
ÆMOTION – A New Approach to Urban Mobility. Aemotion introduces a unique electric vehicle that blends the functionality of a two-wheeler with the safety and comfort features typically found in enclosed vehicles. Designed for agility and convenience, it includes removable
0
0
2
Many such cases in Europe as well.
“The VC eventually understood once I explained why founders can't share secrets with their competitors' mentors.”
1
0
1
Space plumbing FTW!.
Astronautics propulsion often comes down to plumbing, and interstellar propulsion will be no exception. So we had @Swagelok Québec bring their training trailer to campus and put our personnel through a valve & fitting bootcamp.
0
0
1
RT @IgorCarron: 🚀 By crafting a model that can be re-used for further training, @LightOnIO empowers communities with new SOTA capabilities.….
0
3
0
🚀 By crafting a model that can be re-used for further training, @LightOnIO empowers communities with new SOTA capabilities. Today, it is for clinical tasks. Which community will produce the next SOTA model in their field?. I don't know, but I know it will be ✨fabulous✨!.
You can just continue pre-train things ✨.Happy to announce the release of BioClinical ModernBERT, a ModernBERT model whose pre-training has been continued on medical data.The result: SOTA performance on various medical tasks with long context support and ModernBERT efficiency
0
3
13
RT @LightOnIO: 🚀Announcing BioClinical ModernBERT, a SOTA encoder for healthcare AI, developed by Thomas Sounack @tsounack for Dana-Farber….
0
9
0
RT @antoine_chaffin: @jeremyphoward Thanks!.Kudos to @tsounack, he is the driving force behind this project!.
0
1
0