Aurélien Lac

AI Research Engineer @liquidai

[Aurélien Lac @Aurelien_L_]

⚡ A lot of representation models still start from a massive generative VLM.

We trained an encoder from scratch instead: native on text and images, multilingual, and built to be fast from the beginning.

I’m super proud to share NeoMME with my friend @tonywu_71. 260M and 800M, no vision tower. We also fine-tuned a visual document retriever on top: NeoMME-Retriever.

[Aurélien Lac @Aurelien_L_]

@maximelabonne @tugot17 https://t.co/A54UhkNtf2

[Aurélien Lac @Aurelien_L_]

Wake up. Get challenged by @maximelabonne. Get dressed to grab my morning croissant. Win. 🥐

Great start to the weekend

@tugot17 show us what you got

[Maxime Labonne @maximelabonne]

45%, it's a well-calibrated decision model

@Aurelien_L_ it's your turn 💅

[Vercel Developers @vercel_dev]

d1 from @liquidai is live on AI Gateway.

𝚕𝚒𝚚𝚞𝚒𝚍/𝚍𝟷

Make decisions with vision support to classify, route, and score.
https://t.co/425l8tVzhE

[Aurélien Lac @Aurelien_L_]

The kind of release that leaves no choice but to use these for OCR.

Congrats to the @LightOnIO team 👏

[Said Taghadouini @staghado]

🚀 LightOnOCR-3 is out! 🦉

This time we go beyond OCR: text transcription, document layout, image descriptions and chart extraction, all in one model.

We release 3 sizes: 0.8B, 1B and 4B, all under Apache 2.0!

[Piotr Mazurek @tugot17]

also supported in @sgl_project 🫡

[Aurélien Lac @Aurelien_L_]

@antoine_chaffin @helloiamleonie @ESL_Sarah @MaxSchall1 @bo_wangbo @17Ahmetyucel @doesdatmaksense Some days I miss working with you, other days I remember your Slack messages

[Aurélien Lac @Aurelien_L_]

@helloiamleonie Reinforcement Learning from Human Fashion 💃

[Aurélien Lac @Aurelien_L_]

@_jakubchmura @liquidai Glad you like it 🫡

[Aurélien Lac @Aurelien_L_]

@DogukanUrker https://t.co/GT5g1Tig9G

Hope that helps !

[Liquid AI @liquidai]

We built 10 live-camera demos for d1-3B, from gesture-controlled games to content moderation, with one forward pass per frame.

In collaboration with @NVIDIARobotics, we also show d1-3B navigating an environment in Isaac Sim, with the model served on a Jetson in a hardware-in-the-loop setup.

5/

[Aurélien Lac @Aurelien_L_]

@_jakubchmura @liquidai That was fast! Does is already support caching the context for multiple questions ? 👀

[Maxime Labonne @maximelabonne]

Tiny decision models you can run everywhere:

> d1-3B has impeccable performance in text and vision
> d1-omni-600M supports text, vision, and audio!

Try our @huggingface space with interactive games today

[Aurélien Lac @Aurelien_L_]

@antoine_chaffin 🙃

[Aurélien Lac @Aurelien_L_]

I'm always exited to see new strong retrieval models. But this 600M one in particular is very very impressive. Congrats @antoine_chaffin and the team.

[Antoine Chaffin @antoine_chaffin]

New company, new team, same frontier results & same open license
Happy to announce the release of our new embedding model family
More than just an upgrade in performance, the models are now multimodal, multi-vectors, and use a shared embedding space to allow cross-model querying

[Aurélien Lac @Aurelien_L_]

@antoine_chaffin Thanks 🫡 I'm expecting you to come back to your roots too @antoine_chaffin

[Aurélien Lac @Aurelien_L_]

Most decisions don't need a big model. They need a fast one.

Today we're open sourcing d1-3B, a vision-language decision model, and d1-omni-600M, which takes text, images and audio.

[Liquid AI @liquidai]

Today we release Open d1: two open-weight multimodal models in our d1 decision model family.

> d1-3B: text + vision
> d1-omni-600M: text + image or text + audio
> Real-time decision making anywhere, from data centers such as @nvidia DGX to RTX workstations to Jetson at the edge.

1/

[Ramin @ramin_m_h]

here you go! our first series of open-weight multimodal decision models in collaboration with @nvidia 🤝✨

I am obsessed with our 600m omni model that truly brings in new possibilities to the field processing images, audio and text.

Fine-tune these bad-boys for your system one applications! 🚀

masterful work by our post-training and applied AI work.

[Aurélien Lac @Aurelien_L_]

@tonywu_71 The model is the piece of art 😎

[Aurélien Lac @Aurelien_L_]

Huge thanks to @maximelabonne, @FernandoNetoAi, @EdoardMosca, @helloiamleonie, Tarek and the rest of the team, it was very nice to build and test this with you!

[Aurélien Lac @Aurelien_L_]

Models: https://t.co/jJnmLOPivB and https://t.co/tKye3GXtvc
Blog post: https://t.co/c1piP7xWVr
Arcade: https://t.co/Qtev6HRel9

I want to see what you build with it!