Forget just thinking in words. 🚀 New Era of Multimodal Reasoning🚨 🔍 Imagine While Reasoning in Space with MVoT Multimodal Visualization-of-Thought (MVoT) revolutionizes reasoning by generating visual "thoughts" that transform how AI thinks, reasons, and explains itself.

🚀Excited to co-organize the #ICLR2025 Workshop on Modularity for Collaborative, Decentralized, and Continual Deep Learning (MCDC ICLR 2026). Big thanks to an amazing team and speaker lineup! We're calling for non-archival papers in either short papers (2 pages) or long papers

A very interesting direction! We also had an early exploration in this area, where we enabled VLMs to localize a target object by reasoning over user instructions and then utilized a tool to further localize the object in the image. github.com/OptimalScale/D…

Introducing CTRL, a new framework that trains LLMs to critique via RL without human supervision or distillation, enabling them to supervise stronger models and achieve test-time scaling through iterative critique-revisions. 1/ Paper: arxiv.org/abs/2502.03492 Website:

🚀 Introducing VLM²-Bench! A simple yet essential ability that we use in daily life. But when tackling vision-centric tasks without relying on prior knowledge, can VLMs perform well? 🤔 🔗 Project Page: vlm2-bench.github.io More details below! 👇 (1/n)

How to achieve human-level open-ended machine Theory of Mind? Introducing #AutoToM: a fully automated and open-ended ToM reasoning method combining the flexibility of LLMs with the robustness of Bayesian inverse planning, achieving SOTA results across five benchmarks. 🧵[1/n]

🚀Exciting to see how recent advancements like OpenAI’s O1/O3 & DeepSeek’s R1 are pushing the boundaries! Check out our latest survey on Complex Reasoning with LLMs. Analyzed over 300 papers to explore the progress. Paper: arxiv.org/pdf/2502.17419 Github: github.com/zzli2022/Aweso…

Come to play chess with our diffusion reasoning model here: lichess.org/@/diffusearchv0 by Jiacheng Ye ! Check out our research on diffusion reasoning models (DREAMs) here: ikekonglp.github.io/dreams.html to learn how our discrete diffusion approach enables implicit search capabilities!

🤔 Always wondering if a next-token prediction model is the end of planning and reasoning. 🎯 Now excited to announce our team's latest research on exploring a new paradigm to enhance the planning ability of LLMs with DiffuSearch. 🧵1/7

🤔How can we obtain Long-CoT data for theorem proving? 🚀DeepSeek-R1 utilizes large-scale collected Long-CoT data interleaved with RL training to enhance the performance of large reasoning models. Given the importance of Long-CoT data and the challenges in generating them, our

💡Unlocking Efficient Long-to-Short LLM Reasoning with Model Merging We comprehensively study existing model merging methods on efficient Long-to-Short LLM reasoning tasks, and find their huge potential in the field.

📣🔊 Excited to announce the 2nd AI for Math Workshop at #ICML2025 ICML Conference! 🔍 Workshop details: sites.google.com/view/ai4mathwo… 📜 Submit your pioneering work: sites.google.com/view/ai4mathwo…… 🙋 Reviewer nomination: goo.su/UlL3GJ

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

Dream 7B: A general diffusion language model that happens to excel at planning. Without task-specific training, it outperforms Qwen2.5 7B and LLaMA3 8B on countdown and sudoku problems.

Very cool! We also welcome everyone to check out our large diffusion language model called Dream-7B announced last month. We've open-sourced the checkpoint. Try our demo at: huggingface.co/spaces/multimo… For more details, please refer to our blog: hkunlp.github.io/blog/2025/drea…

What happend after Dream 7B? First, Dream-Coder 7B: A fully open diffusion LLM for code delivering strong performance, trained exclusively on public data. Plus, DreamOn cracks the variable-length generation problem! It enables code infilling that goes beyond a fixed canvas.

📢 Update: Announcing Dream's next-phase development. - Dream-Coder 7B: A fully open diffusion LLM for code delivering strong performance, trained exclusively on public data. - DreamOn: targeting the variable-length generation problem in dLLM!

Dream-Coder, trained entirely on public data, achieves state-of-the-art coding performance among open diffusion code LLMs.

To address variable‑length generation, DreamOn dynamically adjusts masked spans during infilling, expanding or contracting them to precisely match the target length.✅

Follow-up to Dream 7B, now focused on code: Dream-Coder 7B is a diffusion-based code LLM from HKU + Huawei Noah’s Ark, built on Qwen2.5-Coder and 322B open tokens. It replaces autoregressive decoding with denoising-based generation, enabling flexible infilling via DreamOn. A