deepseek-ai / DeepEP
DeepEP: an efficient expert-parallel communication library
See what the GitHub community is most excited about today.
DeepEP: an efficient expert-parallel communication library
cuVS - a library for vector search and clustering on the GPU
Tile primitives for speedy kernels
GPU accelerated decision optimization
RTP-LLM: Alibaba's high-performance LLM inference engine for diverse applications.
[ARCHIVED] Cooperative primitives for CUDA C++. See https://github.com/NVIDIA/cccl
Mirage Persistent Kernel: Compiling LLMs into a MegaKernel
NCCL Tests
LLM training in simple, raw C/CUDA
DeepGEMM: clean and efficient FP8 GEMM kernels with fine-grained scaling
Causal depthwise conv1d in CUDA, with a PyTorch interface
Graphics Processing Units Molecular Dynamics
[ICLR2025, ICML2025, NeurIPS2025 Spotlight] Quantized Attention achieves speedup of 2-5x compared to FlashAttention, without losing end-to-end metrics across language, image, and video models.
Lightning fast differentiable SSIM.
CUDA Kernel Benchmarking Library
Instant neural graphics primitives: lightning fast NeRF and more
RAFT contains fundamental widely-used algorithms and primitives for machine learning and information retrieval. The algorithms are CUDA-accelerated and form building blocks for more easily writing high performance applications.
CUDA Library Samples
Fast CUDA matrix multiplication from scratch