Mark (Coolmark)

@NaderLikeLadder @0xSero Let's make it happen

@0xSero Hey!! How much compute do you need?

Putting out a wish to the universe. I need more compute, if I can get more I will make sure every machine from a small phone to a bootstrapped RTX 3090 node can run frontier intelligence fast with minimal intelligence loss. I have hit page 2 of huggingface, released 3 model family compressions and got GLM-4.7 on a MacBook huggingface.co/0xsero My beast just isn’t enough and I already spent 2k usd on renting GPUs on top of credits provided by Prime intellect and Hotaisle. ——— If you believe in what I do help me get this to Nvidia, maybe they will bless me with the pewter to keep making local AI more accessible 🙏

Amazing work by the Nemotron team. Go try out Nemotron 3 Super today!

@llm_wizard SUPER DAY!

Today marks a SUPER DAY. THAT'S RIGHT. NEMOTRON 3 SUPER JUST DROPPED. Model is: FAST. Model is: SMART. Model is: THE MOST OPEN MODEL WE'VE DONE YET. The team really cooked on this one, folks.

DGX Spark 🤝 @dwarkesh_sp

✨ Thank you to our incredible community of developers and researchers driving AI innovation — one model, one commit, and one breakthrough at a time. 🤗 We’re thrilled to celebrate 50,000+ followers on @HuggingFace. Keep building with us: huggingface.co/nvidia

@rohanpaul_ai Openness is the future!

NVIDIA just released Nemotron 3 Nano, a new 30B hybrid open-source reasoning model! Meant to make multi-agent AI cheaper, faster, and easier to trust. Nemotron 3 Nano claims 4x higher throughput than Nemotron 2 Nano and up to 60% fewer reasoning tokens. Multi-agent AI means multiple LLM-based “workers” cooperate on a task, and the hard part is they waste time talking, lose context, and run up inference cost. Nemotron 3 tries to fix that with a hybrid mixture-of-experts design, where the model routes each token through a small slice of the network instead of firing everything every time. Nano is 30B parameters but only about 3B active at once, and it also supports a 1M-token context window so longer workflows stay consistent. NVIDIA also ships the surrounding “open stack”. - A 3T tokens of datasets, a huge pile of text and task examples used for pretraining, post-training, and reinforcement learning, so people can fine-tune Nemotron for coding, reasoning, and multi-step workflows. - plus NeMo Gym, which is the set of ready-made training “sandboxes” where an agent can practice tasks and get scored, and NeMo RL is the reinforcement learning toolkit that actually runs that feedback-style training loop at scale. - And also evaluation tools which are for checking if the model is behaving well, like measuring quality on tasks and running safety checks, so a team can trust what they are shipping. Nemotron 3 Super and Ultra are expected to be available in the first half of 2026.

We released Nemotron 3, our latest family of open models! >> Try out Nano (30B) on @huggingface: huggingface.co/nvidia/NVIDIA-… >> Or read the read white paper here: research.nvidia.com/labs/nemotron/… Have fun!!

@mervenoyann Open all the way! 🙌

NVIDIA cooked! Nemotron3 Nano is in, along with the training datasets and an agent env library 🔥 > A3B/30B hybrid SSM, 1M context size, built for agentic use > leading in both benchmarks and throughput 🙌🏻 > license enables commercial use Super and Ultra coming soon!

@soumyesinghal @nvidia Let's go openness 🙌

🚀 Nemotron 3 Nano is live! Had a blast post-training this model with a cracked team. Its strong for its size, and highly efficient at inference. And true to @nvidia's open release style: weights (BF16/FP8/base) + training recipes + code + datasets. HF: huggingface.co/collections/nv… Blog + Nano tech report: nvda.ws/48RusVt

@kuchaev @ArtificialAnlys 🙌

🚀 Nemotron 3 Nano 30B-A3B is here! Open weights + open data + open source. AA Intelligence Index: 52 (@ArtificialAnlys ) ✅ 1M‑token context ✅ up to 3.3× higher throughput vs similarly sized open models ✅ stronger reasoning/agentic + chat Details + links in the thread 🧵

@fujikanaeda cooking out some MoE goodness

we released a Nano Nemotron earlier this year, but we knew we could do even better. so, we've been cooking on latest redux & trinity-style release. here's the first one out of the oven: Nemotron 3 Nano.

@fahdmirza CUDA 13.1 is huge!!

🚀 NVIDIA just dropped the biggest CUDA update in 20 years. 👌 CUDA Tile lets you write GPU code without managing threads. 🔥 No more threadIdx calculations. No manual memory juggling. ✌️ I broke it down in 10 minutes with actual code examples 👇 youtu.be/bBclt6AFaH4?si… @Coolmark482 @NVIDIAAI @NVIDIAAIDev #cuda13 #cudatile #cutile

@TheAhmadOsman Excellent experiment

testing a theory about the new algo if this shows up for you: > bookmark if you already follow me > like if you don’t > reply for any other reason (including telling me i shouldn’t be on your timeline lol) appreciate you 🫡

@TheAhmadOsman NeMo NeMo NeMo!!

HUGE > introducing nemo data designer > nvidia’s synthetic data factory for people who are done with “just prompt it lol” > the exact machinery that powered the hundreds of billions of tokens for nemotron > now fully opensourced > when internal teams + external customers ask > “how did you generate this much clean data?” > the answer was always the same: “use data designer” > with it you can build datasets from nothing or remix your own seed data > statistical samplers, llms, correlations, validators, the whole toolbox > key features > statistical + llm + seed-dataset generation > dependency-aware fields (no more nonsense columns) > python/sql/remote validators baked in > llm-as-a-judge scoring > preview mode so you can iterate before you go full blast > plugin system for extending your own modules + loading external ones > what you can do > sample people w/ demographic attributes > generate structured rows w/ correlated fields > define custom models + providers > build python/sql/remote validators > run cli tools to manage providers, models, configs > host everything yourself if your company won’t touch outside infra > install > pip install data-designer > or clone + make install > set your api key > export NVIDIA_API_KEY="…" > or OPENAI_API_KEY="…" > basic usage > from data_designer import samplers, config > dd = DataDesigner() > cfg = DataDesignerConfigBuilder() > example > add a product_category column using a category sampler > add a review column using an llm prompt > preview the dataset > then scale to millions > docs include: quick start, notebooks, column types, validators, model configs, person sampling > apache 2.0 license > basically: your synthetic data pipeline just grew up > and you now have nvidia’s internal data-factory workflow on your laptop NVIDIA has been on fire with their OSS releases these past couple of months

@CobusGreylingZA MoE models go brrrrr

To me at least, it feels like so much NVIDIA is doing goes largely unnoticed... they are building an immense groundswell in terms of Agentic ecosystems and model performance... For instance, today NVIDIA announced that they are expanding their partnership with Mistral AI. The new Mistral 3 frontier open models now run up to 10× faster on NVIDIA GB200 NVL72 GPUs. Mistral AI worked closely with NVIDIA to bring the full Mistral 3 family to market with advanced optimisations. Across the board NVIDIA is laser focussed on improving model inference latency...a crucial bottleneck for production implementations. @Coolmark482 @nvidia

@llm_wizard @ArtificialAnlys openness ftw!

Yappin’ about the new Openness Index from the team at @ArtificialAnlys They cooked with this one. youtube.com/shorts/FPB4lbn…

@shizhediao 🙌

Exciting to see enterprises adopting specialized, modular AI systems. This is exactly where the next wave of efficiency and innovation will come from. We will also release new research about the compound agent system and demonstrate its superiority by moving from monolithic models to multi-specialist agent systems. Stay tuned!

@MariyaSha888 Latvia?

Account based in Canada... but not for much longer! 😉 Where do you think I'm moving?

@TheAhmadOsman Yessir!

who here would like to see a build video guide for multiple RTX PRO 6000s? already got the hardware ordered for a couple of 3090s and 5090s build guides btw yes, there'll be GPU giveaways ;) first video guide before Thanksgiving anyway, Buy a GPU keeps on winning

@JacksonAtkinsX @NVIDIAAI Open source vision models FTW!

.@NVIDIAAI just cut video processing tokens 4x. NVIDIA Nemotron Nano 2 VL 12B just dropped. It's the successor to their 3M+ download v1. Major upgrades across the board: - DocVQA: 91.2% → 94.4% - MMMU: 48.2% → 68.0% - Video-MME: 54.7% → 65.9% It's free on Hugging Face.