NVIDIA / cuopt
NVIDIA cuOpt is an open-source GPU-accelerated optimization engine delivering near real-time solutions for complex decision-making challenges.
See what the GitHub community is most excited about today.
NVIDIA cuOpt is an open-source GPU-accelerated optimization engine delivering near real-time solutions for complex decision-making challenges.
CUDA accelerated rasterization of gaussian splatting
FlashInfer: Kernel Library for LLM Serving
cuVS - a library for vector search and clustering on the GPU
NCCL Tests
CUDA Library Samples
DeepEP: an efficient expert-parallel communication library
A massively parallel, optimal functional runtime in Rust
CUDA Kernel Benchmarking Library
LLM training in simple, raw C/CUDA
FlashMLA: Efficient MLA decoding kernels
Causal depthwise conv1d in CUDA, with a PyTorch interface
Quantized Attention achieves speedup of 2-5x and 3-11x compared to FlashAttention and xformers, without lossing end-to-end metrics across language, image, and video models.
Instant neural graphics primitives: lightning fast NeRF and more
[ARCHIVED] Cooperative primitives for CUDA C++. See https://github.com/NVIDIA/cccl