This repository, NVlabs/tiny-cuda-nn, is a high-performance C++/CUDA library and PyTorch extension for training and querying neural networks with a focus on extreme GPU efficiency. It provides a modular, JSON-configurable framework of specialized components: input encodings (multiresolution hash grids, spherical harmonics, frequency, etc.), network architectures (fully fused MLP using Tensor Cores, CUTLASS-based MLP), loss functions, and optimizers (Adam, Shampoo, Novograd, etc.). The library supports runtime JIT compilation to fuse forward, backward, and loss kernels into single CUDA kernels, yielding up to 5x speedups. It is the engine behind NVIDIA's Instant Neural Graphics Primitives (Instant NGP) and is widely used in NeRF and neural rendering research. The codebase is production‑quality, with extensive use of templates, GPU memory arenas, CUDA graphs, multi‑stream parallelism, and comprehensive CI across many CUDA versions and GPU architectures. Although primarily a library, it includes sample applications and benchmarks that demonstrate its use in image overfitting and throughput measurement. PyTorch bindings via pybind11 make it accessible for Python‑based research workflows.

Alpamayo 1 is a Vision-Language-Action (VLA) model designed for autonomous driving applications. It implements the Alpamayo-R1 architecture that combines a Qwen3-VL backbone for vision-language reasoning with a diffusion-based action expert for trajectory prediction. The system takes multi-camera video and egomotion history as inputs, generates Chain-of-Causation (CoC) reasoning traces, and outputs future trajectories (6.4s horizon at 10Hz). Key components include a unicycle kinematic model for action space conversion, CUDA graph acceleration for efficient repeated inference, and robust geometric transformations. While the repository is noted to receive only limited maintenance as newer versions exist, the current implementation demonstrates strong engineering practices with several reusable components.

The repository provides the official PyTorch implementation of DoRA (Weight-Decomposed Low-Rank Adaptation), a parameter-efficient fine-tuning method. It integrates with Hugging Face PEFT and includes training/evaluation scripts for vision-language and language tasks (e.g., VQA, ScienceQA, commonsense reasoning). The architecture is modular, with core components like DoRA tuners, adapter controllers, and low-rank layers. Entry points include training scripts (finetune.py, train_dora.py), evaluation scripts (evaluate.py, run_llava.py), and serving tools (gradio_web_server.py). External dependencies include PyTorch, Transformers, PEFT, bitsandbytes, and dataset-specific libraries.

Timeloop is a framework for modeling, mapping, and design-space exploration of dense and sparse tensor algebra workloads on hardware accelerators. It combines a polyhedral analysis engine (based on the Integer Set Library, ISL) with a configurable architecture model to evaluate mappings (tiling, permutation, datatype bypass) in terms of energy, latency, and area. The primary application is the `timeloop-mapper` executable, which searches over a customizable mapping space using strategies like exhaustive, random, hybrid, or pruned search. A companion `timeloop-model` evaluates a single mapping, and `timeloop-metrics` computes detailed statistics. The codebase is predominantly C++ with Python scripts for automation and post-processing. It also includes subprojects: **Tenssella** (a polyhedral-based emulation and code generation framework) and **Orojenesis** (a Python tool for deriving operational intensity bounds of fused tensor algorithms). While the core is a robust software tool for accelerator architects, the presence of experimental subdirectories, multiple code generations, and research-oriented features (sparse tensor compression, density models, fused workloads) gives it a hybrid character—part production tool, part research prototype. The architecture is modular, with clear separation between mapping representation, loop analysis, hardware modeling, and search algorithms. Key entry points include `src/applications/mapper/mapper.cpp` and `src/applications/model/model.cpp`. External dependencies include ISL, barvinok, Boost, yaml-cpp, and libconfig.

SpatialClaw is a research framework for spatial reasoning agents — LLM/VLM agents that answer spatial questions about images and video by writing and executing Python code. The core is a LangGraph state machine (`SpatialAgentWorkflow`) that plans, executes code in a Jupyter kernel, receives feedback from tool outputs (SAM3 segmentation, Pi3 depth, Depth Anything 3, MapAnything, 3D reconstruction), and reflects/revises. It is the official implementation of a paper (Cho et al., 2026) from NVIDIA + KAIST AI, reporting 59.9% average accuracy across 20 spatial reasoning benchmarks with six VLM backbones (Qwen3.5/3.6, Gemma4, 26B–397B).