Adi Renduchintala

🚀 Introducing Nemotron-Cascade! 🚀 We’re thrilled to release Nemotron-Cascade, a family of general-purpose reasoning models trained with cascaded, domain-wise reinforcement learning (Cascade RL), delivering best-in-class performance across a wide range of benchmarks. 💻 Coding powerhouse After RL, our 14B model: • Surpasses DeepSeek-R1-0528 (671B) on LiveCodeBench v5/v6/Pro. • Achieves silver-medal performance at IOI 2025 🥈. • Reaches a 43.1% pass@1 on SWE-Bench Verified, and 53.8% with test-time scaling. 🧠 What is Cascade RL? Instead of mixing heterogeneous prompts across domains, Cascade RL trains sequentially, domain by domain, which reduces engineering complexity, mitigates heterogeneous verification latencies, and enables domain-specific curricula and tailored hyperparameter tuning. ✨ Key insight Using RLHF for alignment as a pre-step dramatically boosts complex reasoning—far beyond preference optimization. Subsequent domain-wise RLVR stages rarely hurt the benchmark performance attained in earlier domains and may even improve it, as illustrated in the following figure. 🤗 Models & training data 🔥 👉 huggingface.co/collections/nv… 📄 Technical report with detailed training and data recipes 👉 arxiv.org/pdf/2512.13607

Today, @NVIDIA is launching the open Nemotron 3 model family, starting with Nano (30B-3A), which pushes the frontier of accuracy and inference efficiency with a novel hybrid SSM Mixture of Experts architecture. Super and Ultra are coming in the next few months.

NVIDIA releases Nemotron 3 Nano, a new 30B hybrid reasoning model! 🔥 Nemotron 3 has a 1M context window and the best in class performance for SWE-Bench, reasoning and chat. Run the MoE model locally with 24GB RAM. Guide: docs.unsloth.ai/models/nemotro… GGUF: huggingface.co/unsloth/Nemotr…

@mickel_liu @rajammanabrolu Could you point me to one such model? I’d like to bookmark it

@rendu_a @rajammanabrolu now base models come with chat templates baked in

y'all can't just shove all your post training data into pre/mid then call your RL runs "cold start" smh

We found a new way to get language models to reason. 🤯 No RL, no training, no verifiers, no prompting. ❌ With better sampling, base models can achieve single-shot reasoning on par with (or better than!) GRPO while avoiding its characteristic loss in generation diversity.

@willccbb Well there is a group relative reward normalisation bit..

accidentally said “GRPO” instead of “sequence-level REINFORCE with truncated importance sampling ratio correction and clip-higher” and they kicked me out of SF

@RamonAstudill12 Is this the case for non AI fields as well?

Peer review is at risk of disappearing mainly for reasons unrelated to the rise of bureaucrats to power on the orgs that coordinate/control it, but this is definitely making the situation far worse.

We are excited to release Nvidia-Nemotron-Nano-V2 model! This is a 9B hybrid SSM model with open base model and training data. This model also supports runtime "thinking" budget control. HF collection with base and post trained models: huggingface.co/collections/nv…

We have been hard at work on improving hybrid models! Looking forward to see how mamba hybrid models shape the Reasoning LLM space. I’m Super excited to be a part of this effort.

NVIDIA’s Graduate Fellowship Program is now accepting applications for the 2026–2027 academic year. Selected Ph.D. students receive tuition and stipend coverage up to $60K, plus mentorship and technical support from top NVIDIA researchers during an NVIDIA internship. If you’re advancing work in AI, robotics, computer graphics, autonomous vehicles, healthcare, HPC, or related fields — this is your moment. 📅 Apply by Sept. 15, 2025: nvda.ws/3UsbEpe

@jxmnop @Miles_Brundage Would be nice to see how this method works on open source models for which the base is also available.

@Miles_Brundage theoretically, because the update is so low-rank empirically, because the generations have nothing to do with the training data e.g. I didn’t train the model to output Harry Potter somehow it knew that already

OpenAI hasn’t open-sourced a base model since GPT-2 in 2019. they recently released GPT-OSS, which is reasoning-only... or is it? turns out that underneath the surface, there is still a strong base model. so we extracted it. introducing gpt-oss-20b-base 🧵

@sleeping4cat @abeirami Agreed. I think more should be done for recognition of reviewers though. We need to give them some of the spotlight

@abeirami @rendu_a free registration and travel support can go miles.

Instead of complaining that peer review is dead, take a positive step to improve it today. The reviewers are not aliens, they are us! - Revise your review and make it clear. Identify the crucial points that impacted your score negatively and positively. - If the paper is lacking information about its claims, communicate your asks and the reasoning concretely. Don't just ask for 2 more experiments because you feel the authors didn't work hard enough. Don't ask for experiments unless they verify a hypothesis (which you clearly explained). - Look for the missing information that you identified and make sure they are not in the paper. - Recommend acceptance if the paper's claim is adding a new nugget of information to the literature (no matter the size of the nugget), and if the paper has substantiated the claim via theoretical / empirical evidence. Believe me, this doesn't take much time, and will improve the state of peer review significantly!

@abeirami How about invited/panel talks from outstanding reviewers! They can talk about their reviewing process and/or highlight their own research.

@abeirami +1 to all points made. I’d love to figure out ways to incentivize reviews (almost) as much as writing papers. Money? Credits if they are a student?

AI model post training is rapidly improving. The plot below (starting from the same base model) illustrates about 10 months of progress in the *open* post-training research. I’m not convinced that closed research can move as fast.

Transformers are still dominating the LLM scene but we show that higher throughput alternatives exist which are just as strong! Grateful to have a part in Nemotron-H Reasoning effort. 🙏 Technical report will be out soon, stay tuned!

Some people have said that OpenAI achieved state of the art results on the SWE-Bench Verified leaderboard with their codex model, but that's actually not quite correct, no matter how you measure it. A quick 🧵

NSF budgets slashed by 50%, ongoing grants cancelled, NSF staff drastically reduced, all 37 divisions abolished, and grants will now be reviewed by a political kommissar. How will that help technological leadership? linkedin.com/posts/yann-lec…

idk dude I come here and look at my feed, literally everyone on my following feed is subscribed to DOGE and not a single professional scientific researcher has noted that every division at the NSF was just abolished