Curated tools, datasets and references.
Unified CUDA toolkit for Arm on Jetson Thor with full memory coherence, multi-process GPU sharing, OpenRM/dmabuf interoperability, NUMA support, and better tooling across embedded and server-class targets.
Leverage GPU memory swap (model hot-swapping) to share GPUs across multiple LLMs, reduce idle GPU costs, and improve autoscaling while meeting SLAs.
Introduces nvMatmulHeuristics to quickly select a small set of high-potential GEMM kernel configurations for CUTLASS 4.2, drastically reducing auto-tuning time while approaching exhaustive-search performance.
Learn how PyTorch AoT compilation speeds up ZeroGPU Spaces by exporting a compiled model once and reloading instantly, with FP8 quantization, dynamic shapes, and careful integration with the Spaces GPU workflow.
Guide to fine-tuning gpt-oss with SFT + QAT to recover FP4 accuracy while preserving efficiency, including upcasting to BF16, MXFP4, NVFP4, and deployment with TensorRT-LLM.
Explores how small language models enable cost-effective, flexible agentic AI alongside LLMs, with NVIDIA NeMo and Nemotron Nano 2.
A production-ready, modular telesurgery workflow from NVIDIA Isaac for Healthcare unifies simulation and clinical deployment across a low-latency, three-computer architecture. It covers video/sensor streaming, robot control, haptics, and simulation to support training and remote procedures.
CUDA 13.0 introduces shared memory register spilling to reduce local memory pressure by directing spilled registers to on‑chip shared memory when space permits. Opt-in via PTX inline assembly after the function declaration; gains typically 5–10% in register‑pressure workloads.
Guidance on deploying and scaling LangGraph-based agents in production using the NeMo Agent Toolkit, load testing, and phased rollout for hundreds to thousands of users.
Overview of NVIDIA's co-packaged optics (CPO) platform, highlighting photonics/electronics integration, Micro Ring Modulator, COUPE-based engines, and ecosystem collaboration with TSMC and partners for high-density, energy-efficient data-center interconnects.
Jetson Thor combines edge AI compute, MIG virtualization, and multimodal sensors for flexible, real-time robotics at the edge, with FP4/FP8 acceleration and support for Isaac GR00T and large language/vision models.
NVFP4 is a 4-bit data format delivering FP16-level accuracy with the throughput and memory efficiency of 4-bit precision, extended to pretraining for large language models. This profile covers 12B-scale experiments, stability, and industry collaborations.
A developer-focused resource outlining five common pandas bottlenecks, practical CPU and GPU fixes, and drop-in GPU acceleration with cudf.pandas for scalable data workflows.
An in‑depth profile of NVIDIA Blackwell Ultra, its dual‑die NV‑HBI design, NVFP4 precision, 288 GB HBM3e per GPU, and system‑level interconnects powering AI factories and large‑scale inference.
Overview of how NVIDIA's Blackwell hardware, open source models, and a rich software stack accelerate AI—from data prep to deployment—through broad model/tool availability and end-to-end infrastructure.
cuPQC v0.4 expands hash function support and adds Merkle-tree calculations to accelerate data integrity, verification, and privacy-preserving cryptography on GPUs.
Overview of NVLink Fusion, enabling scalable, high-bandwidth AI inference across custom CPU/XPU configurations and rack-scale NVLink fabrics.
NVIDIA HPC SDK v25.7 advances unified memory programming to automate CPU-GPU data movement, boost developer productivity, and simplify porting of ocean models like NEMO to GPUs on Grace Hopper–based systems.
Guidance for deploying and streaming Omniverse Kit Apps at scale with flexible options across cloud, on-prem, and DGX Cloud.
Overview of NeMo-RL v0.3 with Megatron-Core backend for post-training large models, detailing 6D/4D parallelism, GPU-optimized kernels, and simplified configuration to boost reinforcement learning throughput on models at scale.
Learn how to connect Claude to Hugging Face Spaces via the MCP Server to generate images with Krea and Qwen-Image, leverage free credits, and explore the Hugging Face AI App Directory.
Open Nemotron Nano 2 9B delivers leading accuracy and up to 6x throughput with a Hybrid Transformer–Mamba backbone and a configurable thinking budget, aimed at edge, PC and enterprise AI agents.
Overview of NVIDIA's Gamescom 2025 updates: DLSS 4, RTX neural rendering, RTX Kit, NvRTX, ACE AI, and related tooling for scalable, AI-driven gaming experiences.
A practical guide to developing, building for multiple architectures, and deploying CUDA kernels with Hugging Face Kernel Builder. Learn how to create a robust workflow from local development to Hub-based distribution.
A practical walkthrough of Hugging Face's kernel-builder for developing, compiling, and deploying production-ready CUDA kernels across PyTorch, with reproducible builds, multi-arch support, and hub-based distribution.
Explains the Model Context Protocol (MCP) for research discovery and how AI can orchestrate research tools across arXiv, GitHub, and Hugging Face via natural language.
Intro to LangChain with quickstart and examples.
TextQuests is a benchmark testing LLM agents in 25 classic text-based Infocom games, emphasizing long-context reasoning and autonomous exploration.
FilBench is a comprehensive evaluation suite to assess LLM capabilities for Tagalog, Filipino, and Cebuano across cultural knowledge, NLP, reading comprehension, and generation, using a rigorous, historically informed methodology.
A concise resource on composing multiple parallelism strategies (DP, FSDP, TP, CP) with Accelerate and Axolotl to train large models across many GPUs, with guidance on configuration, use cases, and trade‑offs.
AI Sheets is an open-source, no-code tool for building, enriching, and transforming datasets with AI models. Deployable locally or on the Hub, it supports thousands of open models and lets you iterate with prompts, few-shot feedback, and model comparisons.
Overview of TRL's multimodal alignment methods for vision-language models: GRPO, GSPO, MPO, plus native SFT support and DPO-based baselines.
Two open-weight OpenAI models (gpt-oss-120b and gpt-oss-20b) using mixture-of-experts and MXFP4 quantization for fast, memory-efficient reasoning. Licensed Apache 2.0; designed for private/local deployments and on-device use, with Hugging Face Inference Providers integration.
Explore how Gradio MCP servers connect LLMs to Hugging Face models, enabling an AI shopping assistant that browses stores and performs virtual try-ons using IDM-VTON.
Open-source, free Python library for experiment tracking with a local Gradio dashboard that can sync to Hugging Face Spaces for easy sharing and collaboration. Drop-in replacement for wandb with a lightweight beta design.
Overview of the renamed Hugging Face CLI, its reorganized resource-based commands, and the new Jobs service, plus setup, usage, and migration notes.
Enable Parquet CDC with Hugging Face's Xet storage to deduplicate data chunks across files and repositories, reducing upload/download size and storage costs.
A concise resource outlining why multimodal, scale-driven approaches are unlikely to yield human-level AGI and why embodied world models are essential.
Recent advances in Large Language Models (LLMs) enable exciting LLM-integrated applications. However, as LLMs have improved, so have the attacks against them. Prompt injection attack is listed as the #1 threat by OWASP to LLM-integrated applications, where an LLM input contains a trusted prompt (ins
PLAID jointly generates protein 1D sequences and 3D structures by learning the latent space of protein folding models. It enables function- and organism-guided prompts and decodes structure with frozen folding-model weights.
100 RL-controlled cars deployed on I-24 during rush hour to dampen stop-and-go waves, improve throughput, and reduce fuel use for all road users. Decentralized controllers rely on basic radar sensors and local observations.
Explores how mathematics remains central to ML, but its role is evolving from theory-first guarantees to geometry, symmetries, and post-hoc explanations in scale-driven AI.
A method to steer LLMs toward representative, consistent virtual personas by generating naturalistic backstories and using them as conditioning context, enabling individualized simulations and scalable user studies.
Analysis of how ChatGPT responds to different English dialects, highlighting biases against non-standard varieties and implications for global users.
Explores purposeful dialogue in LLM chatbots, arguing multi-turn interactions better align AI with user goals and enable collaboration, especially in coding and personal assistant use cases.
Overview of a high-quality jailbreak benchmark with dual automated evaluators, a 313-prompt dataset, and findings that many jailbreaks underperform claims from earlier work.
A wellbeing-centered framework for beneficial AI that blends psychology, economics, and governance to outline plausible, actionable deployment visions that support individual and societal flourishing.
Benchmark for long-context visual reasoning across large, uncorrelated image sets; introduces MIRAGE to extend LMMs beyond single-image VQA.
A study demonstrating how small language models can perform accurate function calling at the edge using a curated data pipeline, an LLMCompiler-based planner, and on-device execution with macOS integrations.
Overview of how LLMs can be applied to financial markets, including autoregressive modeling of price data, multi-modal inputs, residualization, synthetic data, and multi-horizon predictions, with caveats about market efficiency.
Survey of key work measuring gender bias in AI, across word embeddings, coreference, facial recognition, QA benchmarks, and image generation; discusses mitigation, gaps, and the need for robust auditing.
A deep dive into Mamba, a State Space Model (SSM) based backbone designed for long-context sequences, offering Transformer-like performance with improved efficiency.
End-to-end modeling of extremely large images on contemporary GPUs via nested tokenization and region/context encoders, delivering richer context with lower memory footprints.
Directory of BAIR Lab PhD graduates featuring research interests, advisor(s), and contact details to facilitate collaboration and recruitment.
Survey of how large language models (LLMs) could augment autonomous driving across perception, planning, and generation, with examples, challenges, and early results.
Examines whether text embeddings can be inverted to recover original text, detailing vec2text and its impact on privacy in embedding-based systems.
An in-depth look at common reasons machine learning models fail in real-world data, with practical guidance and a path to safer, more reliable evaluation.
Overview of how deep learning enables single-cell sequencing (sc-seq) insights, including autoencoders for dimensionality reduction, multimodal data integration, and spatial transcriptomics.
Explores the sociotechnical challenges of fish counting at hydroelectric dams and how human expertise integrates with machine learning in regulated environments.
Explores how neural networks can capture classical computation (e.g., Bellman–Ford) via graph neural networks and the idea of algorithmic alignment to improve learning to execute algorithms.
A substantial critique of AI alignment debates, arguing that current alignment work is misaligned with real-world harms and driven by product incentives. Examines RLHF/RLAIF, intent alignment, and the HH H preference model as described in The Gradient essay.
Overview of AI consciousness debates, definitions, and moral status implications. Highlights Block's distinctions (self-, monitoring-, access-, p-consciousness) and Chalmers' easy vs hard problems; discusses need for cross-disciplinary work.