🔥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

Thrilled to share my first project at NVIDIA! ✨ Today’s language models are pre-trained on vast and chaotic Internet texts, but these texts are unstructured and poorly understood. We propose CLIMB — Clustering-based Iterative Data Mixture Bootstrapping — a fully automated

[Approach] ➤ Embeds and clusters web-scale data semantically. ➤ Searches, iteratively and efficiently, for optimal data mixtures using a lightweight proxy model + predictor loop. ➤ Learns how different domains interact, and how the right mix can unlock downstream performance

Decision: Tweet Comment: Okay, here is the summary of this Summary: Summary: Besides this picture, this message hallucinates the full name of our approach based on the acronym, which includes 2 words that appeared ZERO times in the entire paper. ICML Conference

🚀 Can we cast reward modeling as a reasoning task? 📖 Introducing our new paper: RM-R1: Reward Modeling as Reasoning 📑 Paper: arxiv.org/pdf/2505.02387 💻 Code: github.com/RM-R1-UIUC/RM-… Inspired by recent advances of long chain-of-thought (CoT) on reasoning-intensive tasks, we

We are extremely excited to announce mCLM, a Modular Chemical Language Model that is friendly to automatable block-based chemistry and mimics bilingual speakers by “code-switching” between functional molecular modules and natural language descriptions of the functions. 1/2

Soft Soft Soft 🍰

Meet Devstral, our SOTA open model designed specifically for coding agents and developed with All Hands AI mistral.ai/news/devstral

Thrilled to share that our paper has been accepted to #ACL2025 Main 🇦🇹 Huge thanks to my amazing collaborators and my advisor Heng Ji 🙃 📄arxiv.org/abs/2502.17793 Happy to chat about our work as well as MLLM research projects 🙌

Introducing AceReason-Nemotron: Advancing math and code reasoning through reinforcement learning (RL) We propose conducting RL on math-only prompts first, then on code-only prompts. Our key findings include: - Math-only RL significantly boosts both math and code benchmarks! -

Does RL truly expand a model’s reasoning🧠capabilities? Contrary to recent claims, the answer is yes—if you push RL training long enough! Introducing ProRL 😎, a novel training recipe that scales RL to >2k steps, empowering the world’s leading 1.5B reasoning model💥and offering

📢We conduct a systematic study to demystify the synergy between SFT and RL for reasoning models. The result? We trained a 7B model - AceReason-Nemotron-1.1, significantly improved from version 1.0 on math and coding benchmarks. ✅AIME2025 (math): 53.6% -> 64.8% ✅LiveCodeBench

Can we scale 4D pretraining to learn general space-time representations that reconstruct an object from a few views at any time to any view at any other time? Introducing 4D-LRM: a Large Space-Time Reconstruction Model that ... 🔹 Predicts 4D Gaussian primitives directly from

🧠 How can AI evolve from statically 𝘵𝘩𝘪𝘯𝘬𝘪𝘯𝘨 𝘢𝘣𝘰𝘶𝘵 𝘪𝘮𝘢𝘨𝘦𝘴 → dynamically 𝘵𝘩𝘪𝘯𝘬𝘪𝘯𝘨 𝘸𝘪𝘵𝘩 𝘪𝘮𝘢𝘨𝘦𝘴 as cognitive workspaces, similar to the human mental sketchpad? 🔍 What’s the 𝗿𝗲𝘀𝗲𝗮𝗿𝗰𝗵 𝗿𝗼𝗮𝗱𝗺𝗮𝗽 from tool-use → programmatic

😂

Learning to perceive while learning to reason! We introduce PAPO: Perception-Aware Policy Optimization, a direct upgrade to GRPO for multimodal reasoning. PAPO relies on internal supervision signals. No extra annotations, reward models, or teacher models needed. 🧵1/3