Hritik Bansal

📢Scaling test-time compute via generative verification (GenRM) is an emerging paradigm and shown to be more efficient than self-consistency (SC) for reasoning. But, such claims are misleading☠️ Our compute-matched analysis shows that SC outperforms GenRM across most budgets! 🧵

LaViDa: A Large Diffusion Language Model for Multimodal Understanding "We introduce LaViDa, a family of VLMs built on DMs. We build LaViDa by equipping DMs with a vision encoder and jointly fine-tune the combined parts for multimodal instruction following. " "LaViDa achieves…

🙌 We've released the full version of our paper, OpenVLThinker: Complex Vision-Language Reasoning via Iterative SFT-RL Cycles Our OpenVLThinker-v1.2 is trained through three lightweight SFT → RL cycles, where SFT first “highlights” reasoning behaviors and RL then explores and…

📢📢📢 Releasing OpenThinker3-1.5B, the top-performing SFT-only model at the 1B scale! 🚀 OpenThinker3-1.5B is a smaller version of our previous 7B model, trained on the same OpenThoughts3-1.2M dataset.

Excited to share that I will join @Meta FAIR (Seattle 🗻) for my final summer internship w/ @ramakanth1729! 🧑‍🎓Looking forward to meeting new people, learning new things, and chatting about data, algorithms, and evaluation for LLM/VLM reasoning.

🥳 Excited to share that VideoPhy-2 has been awarded 🏆 Best Paper at the World Models Workshop (physical-world-modeling.github.io) #ICML2025! Looking forward to presenting it as a contributed talk at the workshop! 😃 w/ @clarkipeng @YonatanBitton Roman @adityagrover_ @kaiwei_chang…

🚨 New work: LLMs still struggle at Event Detection due to poor long-context reasoning and inability to follow task constraints, causing precision and recall errors. We introduce DiCoRe — a lightweight 3-stage Divergent-Convergent reasoning framework to fix this.🧵📷 (1/N)

Announcing OpenThinker3-7B, the new SOTA open-data 7B reasoning model: improving over DeepSeek-R1-Distill-Qwen-7B by 33% on average over code, science, and math evals. We also release our dataset, OpenThoughts3-1.2M, which is the best open reasoning dataset across all data…

🧑‍🍳Very excited to present LaViDa, one of the first diffusion language models for multimodal understanding! 🌟Unlike autoregressive LMs, you can control the speed-quality tradeoff, and solve constrained generation problems out of the box 📦 🌟 We also release LaViDa-Reason, a…

Great to see that the latest #GeminiDiffusion release benchmarks on our challenging general-purpose reasoning Big Bench Extra Hard dataset! It is now available on HF 🤗: huggingface.co/datasets/BBEH/… Eval code: github.com/google-deepmin…

📢 Submit your cool ideas as short or long papers to the first workshop on the foundations of long video generation, understanding and evaluation 🚀 ramoscsv.github.io/longvid_founda…