🔥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

Our paper has been accepted to ICLR 2026 #ICLR2025 We hope this first 7B diffusion language model inspires the community to further explore the potential of diffusion models. Let’s push the boundaries together! 🚀 Code: github.com/HKUNLP/DiffuLL…

It's fascinating to experience firsthand how quickly we can reproduce some patterns from R1-Zero (RL starts directly from the base model!) with such minimal resources 👀

very cool that model as small as 1.5B can emerge this self-verification behavior with RL

Introducing SWE-Gym: An Open Environment for Training Software Engineering Agents & Verifiers Using SWE-Gym, our agents + verifiers reach new open SOTA - 32%/26% on SWE-Bench Verified/Lite, showing strong scaling with more train / test compute arxiv.org/abs/2412.21139 [🧵]

Text Diffusion and AR LLM might not be substitutes for each other but rather complementary, enhancing each other. Mercury has taken a solid step forward.

Check out our recent work towards enhancing controllability and strategic planning by integrating “macro actions” to improve emotionally intelligent conversations in chatbots! linkedin.com/feed/update/ur…

🎉 Milestone Release! AReaL-boba, our latest #RL system! github.com/inclusionAI/AR… #AI • data/code/model ALL🔥 #OPENSOURCE • Full #SGLang & 1.5x faster on 7B RL • SOTA 7B math reasoning: 61.9 AIME24 & 48.3 AIME25 • 200-sample 32B tuning match QwQ on AIME24 Qwen 1/3 👇

🚀Excited to announce Dream 7B (Diffusion reasoning model): the most powerful open diffusion large language model to date.

Due to some conflicts of time, our poster is rescheduled to Saturday afternoon 3-5:30 PM. Hall 3 + Hall 2B #625 Please stop by if you are interested! Thanks,

Proud to report that TarFlow is accepted to #ICML2025 as a Spotlight 🎉 I’m really looking forward to new ideas and applications enabled by powerful Normalizing Flow models 🚀