Knowledge catalogue
Bringing together the world’s brightest minds and the latest accelerated computing technology leads to powerful breakthroughs that help tackle some of the biggest research problems. To foster such inn
I'm part of the Lokutor team that built this. Model: NVIDIA Conformer-CTC Small (13M params, int8). It runs on an ESP32-S3 with 8 MB PSRAM, no GPU or NPU. LibriSpeech WER is 3.7 / 8.2, versus 6.3 / 15
Taiwan-based neocloud GMI Cloud Inc. wants to capitalize on what it says is an “unprecedented” demand for artificial intelligence computing power after raising a massive 663 million in funding from tw
arXiv:2609.36413v1 Announce Type: new Abstract: A pretrained video world model admits many plausible futures for a scene, but a robot must realize the exact task-conditioned one. To turn world models
arXiv:2609.38095v1 Announce Type: new Abstract: Backpropagation (BP) dominates deep learning but imposes a massive memory tax. For example, training OPT-30B with Adam requires approx 600GB of GPU memo
arXiv:2609.37850v1 Announce Type: new Abstract: Large generative models can recover realistic detail in real-world video super-resolution (VSR), but processing an entire video with them is computation
So I would like to share my analysis of PD disagg that I've run by Horace and many folks at Thinky You may have heard that PD disagg helps with per-token 'tail latencies' in ITL/TPOT, as prefill inter
Wall Street Journal: Sources: Jensen Huang and other execs asked Dario Amodei at the White House why he was so extreme in public on AI risks; Amodei said it's important to be honest — Nvidia's CEO was
Stealthium has joined the Vultr Cloud Alliance, adding GPU‑level observability and runtime security to Vultr’s AI infrastructure platform. As a member of NVIDIA’s Inception Program, Stealthium supplie
arXiv:2609.38169v1 Announce Type: cross Abstract: Linear attention replaces growing KV caches with fixed-size recurrent states, yet these persistent states can become a substantial memory bottleneck u
this should be fraud. certainly taxpayers shouldn’t be subsidizing Meta’s data centers. 🦔Meta classified its AI data centers as 'experimental pilot facilities' on its tax return. That let the Nvidia c
Top tech executives met with President Donald Trump on Tuesday to discuss the future of artificial intelligence amid growing concerns over AI models losing control. Researchers and lawmakers are calli
NVIDIA NeMo Relay supplies a common observability layer that captures ordered lifecycle events and structured trajectories—ATOF event streams, ATIF step‑by‑step footprints, and OpenTelemetry spans wit
arXiv:2609.37241v1 Announce Type: new Abstract: User-written Triton kernels enable high-performance GPU computation within PyTorch, but their end-to-end latency can remain dominated by host-side orche
Your agent got the right answer. How much work did it take to get there? We worked with @NousResearch/@Teknium on a hands-on walkthrough of NVIDIA NeMo Relay to collect traces for Hermes Agent. Run tw
Absolutely love this!! Open source all the way!! getting acquired by @nvidia = hugging face can now hire people we couldn't as a small startup and give them a decade to make open-source AI win! if you
arXiv:2609.34034v1 Announce Type: cross Abstract: A central aim of neuromorphic computing is to provide a viable alternative to highly energy-intensive Transformer-based AI. However, efficient alterna
Bloomberg: After lunch with AI leaders, Trump rejects new federal AI rules, says “we automatically have regulation” via DOJ and FBI but “self-regulation is very important” — President Donald Trump sho
NVIDIA TensorRT Model Connect is an open‑source library that delivers AI model reference implementations in C++ built on top of NVIDIA TensorRT, designed to make the inference stack accessible to deve
arXiv:2609.31683v1 Announce Type: new Abstract: This work presents an end-to-end demonstration of autonomous machine learning research conducted by an agent skill on consumer hardware. The demonstrati
arXiv:2609.33536v1 Announce Type: new Abstract: Large language model (LLM) serving is increasingly constrained by the GPU memory consumed by key-value (KV) caches. Existing compression, eviction, and
dream job token maxxing till open source wins getting acquired by @nvidia = hugging face can now hire people we couldn't as a small startup and give them a decade to make open-source AI win! if you're
arXiv:2609.34608v1 Announce Type: new Abstract: World Action Models (WAMs) enable future-aware control by jointly modeling actions and environment dynamics. However, iterative diffusion or flow infere
arXiv:2609.33582v1 Announce Type: new Abstract: Recent real-world image super-resolution (SR) methods often adapt text-to-image (T2I) backbones with ControlNet-style branches or spatial conditioning t
arXiv:2609.34070v1 Announce Type: new Abstract: Long-horizon micromagnetic simulation remains expensive because conventional and learned solvers typically propagate Landau--Lifshitz--Gilbert (LLG) dyn
arXiv:2609.33410v1 Announce Type: new Abstract: Autoregressive decode repeatedly streams a growing KV cache, making attention a major cost at long context. Existing high-performance kernels use online
On September 29, 2026, Hugging Face co‑founder Clement Delangue announced that the company had been acquired by NVIDIA. The acquisition will let Hugging Face scale its hiring beyond what a small start
arXiv:2609.32540v1 Announce Type: new Abstract: Few-step autoregressive video diffusion generates a long video by splitting the video into temporal chunks and generating chunk-by-chunk, each through a
arXiv:2609.32481v1 Announce Type: new Abstract: Joint-embedding predictive architectures are unusually sensitive to how the input is masked: block masks work, scattered masks do not, and the explanati
arXiv:2609.35750v1 Announce Type: new Abstract: Scaling the horizon of agentic LLMs is bottlenecked by the need to fit ever longer context traces in GPU memory. Context compaction has been the most po
arXiv:2609.34060v1 Announce Type: new Abstract: Prompt-specialized multi-agent systems enable multiple agents to share a model while performing complementary roles to solve complex tasks. However, age
arXiv:2609.35665v1 Announce Type: cross Abstract: Interior-point methods (IPMs) are among the most widely used algorithms for constrained optimization, yet their Newton-based search directions require
arXiv:2609.35297v1 Announce Type: new Abstract: Pretraining a language model takes enormous compute, and the right optimizer can save a good part of it. Muon's spectral step gives a stronger direction
NVIDIA released VSS Blueprint 3.3, a framework that stitches together vision‑language models (Cosmos), LLMs (Nemotron) and retrieval‑augmented generation to convert live or recorded video into natural
arXiv:2609.34702v1 Announce Type: new Abstract: Future Mars missions will require rover autonomy that can operate across unstructured terrain, changing illumination, atmospheric dust, and limited comm
arXiv:2510.13602v3 Announce Type: replace-cross Abstract: Decoding throughput improvements from larger inference batches are limited by GPU memory, which is largely consumed by the key-value (KV) cach
Nvidia Corp. is extending the data processing unit from infrastructure offload to a broader security role across the artificial intelligence factory. The opportunity is to make agentic AI safer to ope
Our Secure Intelligence Institute works with researchers at Stanford, CMU, Duke, Columbia, Ohio State, and UVA. We’re also working with NVIDIA and the Open Secure AI Alliance, and sharing tools and fi
arXiv:2609.34133v1 Announce Type: new Abstract: Photographic color editing is inherently personal: the same image can appear too warm, too muted, or already satisfactory to different users. Most looku
arXiv:2609.34555v1 Announce Type: new Abstract: Long-context LLM serving is increasingly bottlenecked by decode, where large KV caches limit batch size and underutilize GPUs. Sparse KV cache offloadin
arXiv:2609.35629v1 Announce Type: new Abstract: Attention often concentrates on a small subset of tokens in the context, but which subset matters changes from one query to the next. To exploit this ch
arXiv:2609.33218v1 Announce Type: new Abstract: Visual world models enable robotic planning by predicting future observations, but dense latent-state propagation and sample-intensive trajectory optimi
arXiv:2609.34562v1 Announce Type: new Abstract: MeMo (Zanzotto et al., 2025) is a recent language-model architecture that stores associations between token contexts and next tokens in a correlation ma
Financial Times: Sources: Nvidia is in early-stage talks with insurance companies to structure risk-mitigation products to protect lenders against loan defaults by neoclouds — World's largest listed c
arXiv:2609.32197v1 Announce Type: cross Abstract: Diffusion Transformers have become the dominant architecture for video generation. Their substantial computational cost motivates scaling inference ac
arXiv:2609.34370v1 Announce Type: new Abstract: Spexis is a multi-GPU LLM inference framework that improves the efficiency of pipeline and tensor parallelism through speculative parallelism. Rather th
arXiv:2609.32672v1 Announce Type: new Abstract: Predicting whether a trained model will generalize under distribution shift remains difficult, especially when target-domain data are unavailable. We in
arXiv:2609.35065v1 Announce Type: cross Abstract: Reusable prefix key-value (KV) caches can outgrow GPU memory in large language model (LLM) serving. A memory-semantic flash hierarchy offers SSD-backe
arXiv:2609.35493v1 Announce Type: new Abstract: Autonomous planetary exploration requires robots to navigate unknown, uneven terrain while assessing risk, traversability, and energetic cost. Quadruped
arXiv:2609.35481v1 Announce Type: cross Abstract: Dynamic routing creates severe load imbalance in large-scale expert-parallel Mixture-of-Experts (MoE) training, turning GPUs that host hot experts int
arXiv:2609.32806v1 Announce Type: new Abstract: Motion planning in robotics requires not only computing collision-free paths but also certifying infeasibility when no such path exists. Complete method
arXiv:2609.32518v1 Announce Type: new Abstract: Distilling bidirectional multi-step video diffusion transformers into few-step causal models has become a common approach for streaming video generation
arXiv:2609.34606v1 Announce Type: new Abstract: Leveraging the paradigm of autoregressive diffusion, text-conditioned interactive video world models aim to simulate temporally coherent environments gu
arXiv:2609.30541v1 Announce Type: new Abstract: Optimizing embedding systems for production recommendation pipelines demands systematic exploration that consumes disproportionate engineering effort at
arXiv:2609.31430v1 Announce Type: new Abstract: Tool observations dominate the context of software-engineering agents, making long interaction histories costly to maintain. Existing context compressio
arXiv:2609.31551v1 Announce Type: cross Abstract: Disaggregating the two stages, Prefill and Decode, onto separate GPU pools is now a standard optimization for (text-only) LLM serving. However, multim
every agent will need a sandbox and the harness should live outside it - separate the brains from the hands great to see nvidia open sourcing tools here, agree with Andrew this is part of the harness
NVIDIA’s DSX MaxLPS employs policy‑governed power sharing to dynamically reallocate power across GPU resources, enabling up to 40 % more GPUs within the same approved power budget. In a joint test at
Other than the obvious self promotion, is there a practical reason people do this that I am missing? There's dozens of llamacpp forks with silly names that are supposedly 'optimized' for this or that
arXiv:2609.30846v1 Announce Type: new Abstract: In this paper, we propose I-Parakeet, an integer-only implementation of NVIDIA's Parakeet-CTC (0.6B parameters) that runs on a smartphone NPU without an