🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

PSA: Stop pretraining your VLMs on EN-filtered data, even if it improves ImageNet and COCO‼️ Doing so impairs the model's understanding of non-English cultures❗️ I argued for years, now finally publish concrete results for this (imo) intuitively obvious recommendation A🧾🧶

This work was carried by Angéline Pouget. She's a hidden gem. She's doing her masters now and will be on the PhD/Industry research market in ~6mo. I hope she'll join us :) All authors made significant contributions. It was a long, intense, fun project. arxiv.org/abs/2405.13777

Embrace cultural diversity in your large-scale data! 🌎🌍🌏 Angéline Pouget’s study shows that (quantitatively) you have no reason not to 🌸

thank you authors, reviewers and speakers for your contributions to a great DMLR ICML Conference presentations by Aditi Raghunathan Stella Biderman @ ICML Lucas Beyer (bl16) Alex Dimakis Nomic AI Matthias Gerstgrasser Angéline Pouget at icml.cc/virtual/2024/w… join & contribute at discord.com/invite/FswYXMv…

🚀🚀PaliGemma 2 is our updated and improved PaliGemma release using the Gemma 2 models and providing new pre-trained checkpoints for the full cross product of {224px,448px,896px} resolutions and {3B,10B,28B} model sizes. 1/7

Attending #NeurIPS2024? If you're interested in multimodal systems, building inclusive & culturally aware models, and how fractals relate to LLMs, we've 3 posters for you. I look forward to presenting them on behalf of our GDM team @ Zurich & collaborators. Details below (1/4)

🔥Excited to introduce RINS - a technique that boosts model performance by recursively applying early layers during inference without increasing model size or training compute flops! Not only does it significantly improve LMs, but also multimodal systems like SigLIP. (1/N)

📢2⃣ Yesterday we released SigLIP 2! TL;DR: Improved high-level semantics, localization, dense features, and multilingual capabilities via drop-in replacement for v1. Bonus: Variants supporting native aspect and variable sequence length. A thread with interesting resources👇

Think you know Gemini? 🤔 Think again. Meet Gemini 2.5: our most intelligent model 💡 The first release is Pro Experimental, which is state-of-the-art across many benchmarks - meaning it can handle complex problems and give more accurate responses. Try it now →

Our post-training team (once again) went beyond to post-craft this fantastic performance-for-size model: Johan Ferret Alexandre Ramé Sarah Perrin Geoffrey Cideron Nino Vieillard Sabela Angéline Pouget V Carbune L Rouillard L Hussenot Gaël Liu Danila Sinopalnikov Olivier Bachem 2/2

Gemma 3n 🚀🚀🚀