Open-source gpu-computing software
386 open-source libraries, tools and applications tagged gpu-computing, ranked by how closely gpu-computing is their primary purpose, then by adoption and maintenance health. Each entry links to a page with sources, use cases and when to choose or avoid it.
386 products, primary matches first, then adoption-weighted; health v2 shown.
| Product | Health v2 | Stars | Maturity |
|---|---|---|---|
| calebwin/emu Emu is a Rust GPGPU library providing a CUDA-like compute-focused abstraction over WebGPU. It offers type-safe GPU data boxing, an auto-man… | 32 | 1606 | maintenance |
| gpushare A Kubernetes scheduler extender that enables multiple pods to share the same NVIDIA GPU device, increasing GPU utilization for inference an… | 32 | 1532 | maintenance |
| mit-han-lab/proxylessnas ProxylessNAS is a neural architecture search (NAS) framework that directly searches CNN architectures on the target task and target hardwar… | 32 | 1448 | maintenance |
| VictorTaelin/WebMonkeys A JavaScript library that lets you spawn thousands of parallel tasks on the GPU through a minimal set/get/work API, using GLSL 1.0 without … | 32 | 1443 | maintenance |
| VerticalResearchGroup/miaow MIAOW is an open source GPU implementation of the AMD Southern Islands ISA written in Verilog. It was developed as a research project at th… | 36 | 1420 | maintenance |
| NVIDIA/VideoProcessingFramework NVIDIA's Video Processing Framework (VPF) provides C++ libraries with Python bindings for fully hardware-accelerated video decoding, encodi… | 10 | 1379 | maintenance |
| anderskm/gputil GPUtil is a Python module that queries NVIDIA GPU status (memory usage, load, availability) by wrapping the nvidia-smi command-line tool. I… | 61 | 1215 | maintenance |
| enazoe/yolo-tensorrt A C++ wrapper around NVIDIA TensorRT for running YOLO object detection models (YOLOv3, YOLOv4, YOLOv5) with support for FP32, FP16, and INT… | 54 | 1201 | maintenance |
| ROCm/MIOpen MIOpen is AMD's open-source library providing high-performance GPU kernels for machine learning primitives (convolutions, normalization, ac… | 89 | 1195 | maintenance |
| Rust-GPU/rust-cuda An ecosystem of Rust libraries and tools for writing and executing fast GPU code fully in Rust, targeting NVIDIA CUDA. It lets developers a… | 66 | 5399 | experimental |
| VoltaML/voltaML VoltaML is a lightweight Python library that compiles and optimizes machine learning and deep learning models for high-performance inferenc… | 32 | 1177 | maintenance |
| punica-ai/punica Punica is a Python system for serving many LoRA-finetuned LLMs from a single copy of the base model on one GPU, using a custom CUDA kernel … | 10 | 1175 | maintenance |
| NVIDIA-Merlin/NVTabular NVTabular is a GPU-accelerated feature engineering and preprocessing library for tabular data, built to handle terabyte-scale datasets for … | 57 | 1151 | maintenance |
| baidu-research/DeepBench DeepBench is a benchmarking suite from Baidu Research that measures the performance of fundamental deep learning operations (matrix multipl… | 32 | 1109 | maintenance |
| GPUOpen-LibrariesAndSDKs/RadeonRays_SDK RadeonRays is a ray intersection acceleration library developed by AMD, providing a C API for scene building and asynchronous ray intersect… | 23 | 1101 | maintenance |
| openai/blocksparse A Python package of TensorFlow ops and custom CUDA GPU kernels for block-sparse matrix multiplication and convolution, plus related ops lik… | 10 | 1067 | maintenance |
| NVlabs/cuda-oxide cuda-oxide is an experimental rustc codegen backend that compiles Rust GPU kernels directly to CUDA PTX, letting you write SIMT device code… | 82 | 3746 | experimental |
| NVIDIA/stdexec stdexec is NVIDIA's reference implementation of std::execution, the C++26 standard framework for asynchronous and parallel programming base… | 77 | 2464 | experimental |
| chelsea0x3b/dfdx A Rust deep learning library providing shape-checked tensors and neural network building blocks with compile-time verification of tensor op… | 23 | 1934 | experimental |
| hughperkins/VeriGPU VeriGPU is an open-source GPU written in SystemVerilog, loosely based on the RISC-V ISA and targeting eventual ASIC tape-out for machine le… | 32 | 1395 | experimental |
| tspeterkim/flash-attention-minimal A minimal ~100-line CUDA re-implementation of the Flash Attention forward pass, loadable as a PyTorch extension. It is designed as an educa… | 26 | 1192 | experimental |
| huggingface/optimum-nvidia Optimum-NVIDIA is a Python library that bridges Hugging Face Transformers with NVIDIA TensorRT-LLM for highly optimized LLM inference on NV… | 62 | 1039 | experimental |
| NervanaSystems/neon neon was Intel Nervana's reference deep learning framework in Python, focused on best-in-class training performance on CPUs, GPUs, and Nerv… | 10 | 3860 | abandoned |
| NVIDIA/libcudacxx NVIDIA's libcudacxx is a C++ Standard Library implementation designed to work across both host CPUs and NVIDIA GPUs, providing standard-con… | 10 | 2300 | abandoned |
| openai/sparse_attention A Python library of sparse attention primitives from OpenAI's Sparse Transformers paper, including fused normal attention and strided/fixed… | 10 | 1612 | abandoned |
| ELS-RD/kernl Kernl is an open-source PyTorch model optimization library that accelerates transformer inference on GPUs with a single line of code, using… | 10 | 1583 | abandoned |
| ivan-vasilev/neuralnetworks A Java library implementing deep learning algorithms and deep neural network architectures (MLPs, CNNs, RBMs, autoencoders, DBNs) with GPU … | 23 | 1236 | abandoned |
| eholk/harlan Harlan is a small domain-specific language implemented in Scheme for programming GPUs, compiled to OpenCL. It serves primarily as a researc… | 32 | 1191 | abandoned |
| hannes-brt/hebel Hebel is a GPU-accelerated deep learning library in Python built on PyCUDA, implementing feed-forward neural networks for classification an… | 23 | 1169 | abandoned |
| antonmks/Alenka Alenka is a GPU-based column-store database engine written in CUDA that leverages vectorized processing and GPU memory bandwidth for fast a… | 32 | 1166 | abandoned |
| NVIDIA/tensorflow NVIDIA's fork of TensorFlow 1.15 providing GPU support for newer NVIDIA hardware and CUDA libraries after Google stopped maintaining the TF… | 10 | 1147 | abandoned |
| dmlc/minpy MinPy is a Python library providing a pure NumPy interface backed by MXNet, automatically dispatching operators to GPU when available and f… | 10 | 1094 | abandoned |
| NVIDIA/gpu-monitoring-tools A deprecated NVIDIA repository containing Go bindings for NVML and DCGM plus the DCGM-Exporter for exposing GPU metrics to Prometheus in Ku… | 10 | 1075 | abandoned |
| NervanaSystems/maxas MaxAs is an assembler for NVIDIA Maxwell GPU architecture that assembles SASS (Shader Assembly) instructions, written in Perl. It enables l… | 10 | 1074 | abandoned |
| GPUOpen-Archive/CodeXL CodeXL is AMD's open-source GPU tool suite offering GPU debugging, profiling, and static kernel/shader analysis for OpenCL, OpenGL, Vulkan,… | 10 | 1011 | abandoned |
| tinygrad/tinygrad tinygrad is a small, hackable deep learning framework providing a PyTorch-like tensor library with autograd, plus a visible IR and compiler… | 88 | 33740 | active |
| karpathy/llm.c llm.c is a pure C/CUDA implementation for training large language models like GPT-2 and GPT-3, with no dependency on PyTorch or Python. It … | 34 | 31176 | active |
| Genesis World Genesis World is a Python simulation platform for physical AI and robotics that combines a unified multi-physics engine, a photorealistic r… | 95 | 30058 | active |
| MLX MLX is an array computation framework for machine learning on Apple silicon, developed by Apple ML research. It offers NumPy-like Python AP… | 95 | 28839 | active |
| PaddlePaddle PaddlePaddle is an industrial-grade deep learning framework written in C++ with Python APIs, supporting high-performance single-machine and… | 92 | 24105 | active |
| NVIDIA/Megatron-LM NVIDIA's GPU-optimized library for training large transformer models at scale, comprising Megatron-LM (reference training scripts) and Mega… | 100 | 18178 | active |
| ggml-org/ggml ggml is a plain C/C++ tensor library for machine learning with no dependencies, designed to be simple, portable, and efficient. It supports… | 99 | 15480 | stable |
| apache/tvm Apache TVM is an open machine learning compiler framework that takes pre-trained models and compiles them into optimized, deployable module… | 95 | 13840 | active |
| LTX-Video Official Python package from Lightricks providing inference pipelines and LoRA training for LTX-2/LTX-2.5, an open-weights DiT-based founda… | 45 | 11138 | active |
| OpenVINO OpenVINO is an open-source toolkit from Intel for optimizing and deploying deep learning inference across CPU, GPU, and NPU hardware. It su… | 95 | 10996 | stable |
| SkyPilot SkyPilot is an open-source AI compute platform that unifies fragmented infrastructure (Kubernetes, Slurm, VMs, 20+ clouds) into a single po… | 100 | 10724 | active |
| facebookresearch/xformers xFormers is a PyTorch-based library of hackable, optimized Transformer building blocks with custom CUDA kernels for fast, memory-efficient … | 87 | 10565 | active |
| deepseek-ai/DeepEP DeepEP is a high-performance GPU communication library for expert parallelism (EP) in MoE training and inference, providing high-throughput… | 55 | 10263 | active |
| OpenRLHF/OpenRLHF OpenRLHF is a high-performance, production-ready open-source RLHF framework built on Ray + vLLM + DeepSpeed for scalable reinforcement lear… | 89 | 10040 | active |
| NVlabs/Sana SANA is an efficiency-oriented PyTorch codebase for high-resolution text-to-image and text-to-video generation built on Linear Diffusion Tr… | 70 | 9129 | active |
| DeepDive slime is an open-source LLM post-training framework for reinforcement learning scaling, connecting Megatron-based training with SGLang-base… | 79 | 8606 | active |
| isaac-sim/IsaacLab Isaac Lab is a GPU-accelerated open-source framework for robot learning built on NVIDIA Isaac Sim, unifying workflows like reinforcement le… | 96 | 8397 | active |
| InternLM/lmdeploy LMDeploy is a toolkit for compressing, quantizing, deploying, and serving large language models, built around its high-performance TurboMin… | 96 | 8107 | active |
| TensorRTx A C++ collection of popular deep learning networks (YOLO variants, ResNet, MobileNet, Swin Transformer, OCR models, and more) implemented f… | 73 | 7838 | active |
| leejet/stable-diffusion.cpp A pure C/C++ inference engine for diffusion models (Stable Diffusion, FLUX, Wan, Qwen Image, Z-Image, and more) built on ggml in the style … | 92 | 7734 | active |
| EmbarkStudios/rust-gpu rust-gpu is a compiler toolchain that makes Rust a first-class language for writing GPU graphics and compute shaders, compiling Rust code t… | 10 | 7557 | active |
| EleutherAI/gpt-neox GPT-NeoX is EleutherAI's library for training large-scale autoregressive transformer language models on GPUs, built on NVIDIA's Megatron an… | 67 | 7471 | active |
| vLLM vLLM-Omni is a Python framework extending vLLM for efficient inference and serving of omni-modality models, including diffusion transformer… | 84 | 7280 | active |
| NVIDIA/warp NVIDIA Warp is a Python framework that JIT-compiles regular Python functions into efficient CPU or CUDA GPU kernels. It provides differenti… | 99 | 7215 | active |
| kvcache-ai/Mooncake Mooncake is a KVCache-centric disaggregated serving platform for LLM inference, originally built to serve Kimi by Moonshot AI. It separates… | 90 | 6835 | active |
| halide/Halide Halide is an embedded DSL (in C++ and Python) for writing high-performance, data-parallel image and array processing pipelines. It separate… | 68 | 6620 | stable |
| fla-org/flash-linear-attention A PyTorch library providing hardware-efficient implementations of emerging sequence model architectures, including linear attention, sparse… | 86 | 5940 | active |
| areal-project/AReaL AReaL is a large-scale asynchronous reinforcement learning system that bridges foundation model training with agent-based applications, sup… | 86 | 5842 | active |
| pytorch/torchtitan torchtitan is a PyTorch-native platform for large-scale training of generative AI models, offering a clean-room implementation of PyTorch's… | 81 | 5830 | active |
| AidLux (AidLearning) AidLux (originally AidLearning) is an AIoT development platform that runs a native Ubuntu Linux environment with GUI, deep learning tooling… | 63 | 5812 | active |
| NVIDIA/DALI NVIDIA DALI is a GPU-accelerated data loading and preprocessing library with optimized building blocks and an execution engine for deep lea… | 91 | 5761 | active |
| pjreddie/darknet Darknet is an open-source neural network framework written in C and CUDA, best known as the original home of the YOLO real-time object dete… | 32 | 26503 | maintenance |
| nerfstudio gsplat is an open-source Python library with CUDA-accelerated, differentiable rasterization of Gaussians, based on 3D Gaussian Splatting fo… | 73 | 5716 | active |
| mosaicml/composer Composer is an open-source PyTorch-based deep learning training library by MosaicML (now Databricks) for training neural networks faster an… | 66 | 5499 | active |
| RLinf/RLinf RLinf is an open-source, flexible and scalable reinforcement learning training infrastructure for embodied AI (vision-language-action model… | 72 | 5453 | active |
| NVIDIA/Model-Optimizer NVIDIA Model Optimizer (ModelOpt) is a Python library of state-of-the-art model optimization techniques including quantization, pruning, di… | 92 | 5386 | active |
| lightvector/KataGo KataGo is an open-source Go (baduk) engine trained via AlphaZero-like self-play, one of the strongest Go bots available. It runs as a GTP e… | 96 | 5204 | active |
| hao-ai-lab/FastVideo FastVideo is a unified Python framework for post-training and real-time inference of video diffusion models, covering data preprocessing, f… | 79 | 4616 | active |
| mosaicml/llm-foundry LLM Foundry is a PyTorch-based codebase for training, finetuning, evaluating, and deploying large language models from 125M to 70B+ paramet… | 62 | 4449 | active |
| FedML FedML (TensorOpera) is a unified Python library for large-scale distributed training, model serving, and federated learning across GPU clou… | 41 | 4070 | active |
| IREE IREE is an MLIR-based end-to-end machine learning compiler and runtime that lowers models from frameworks like PyTorch, TensorFlow, JAX, an… | 95 | 3980 | active |
| Nunchaku Nunchaku is a high-performance inference engine for 4-bit quantized diffusion models (and LLMs) based on the SVDQuant technique from an ICL… | 73 | 3953 | active |
| vllm-project/llm-compressor LLM Compressor is a Python library for applying quantization and pruning algorithms to large language models, producing compressed-tensors … | 92 | 3886 | active |
| predibase/lorax LoRAX is a multi-LoRA inference server that serves thousands of fine-tuned LLM adapters on a single GPU by sharing a common base model. It … | 60 | 3835 | active |
| NVIDIA/skills A catalog of NVIDIA-verified Agent Skills — portable instruction sets that teach AI coding agents like Claude Code and Codex how to use NVI… | 60 | 3668 | active |
| NVIDIA/TransformerEngine Transformer Engine is an NVIDIA library for accelerating Transformer model training and inference on NVIDIA GPUs using low-precision format… | 99 | 3576 | active |
| huggingface/optimum Optimum is a Hugging Face library that extends Transformers, Diffusers, timm, and Sentence Transformers with hardware-specific optimization… | 91 | 3503 | active |
| alibaba/ROLL ROLL is an open-source reinforcement learning library from Alibaba for training large language models at scale, supporting algorithms like … | 82 | 3429 | active |
| LeelaChessZero/lc0 Lc0 is an open-source, UCI-compliant chess engine that plays chess using neural networks trained via AlphaZero-style self-play reinforcemen… | 74 | 3248 | active |
| determined-ai/determined Determined is an open-source deep learning platform that combines distributed training, hyperparameter tuning, experiment tracking, and GPU… | 36 | 3245 | active |
| Jittor/jittor Jittor is a high-performance deep learning framework from Tsinghua University based on just-in-time (JIT) compilation and meta-operators, w… | 68 | 3237 | active |
| Miles Miles is an open-source, enterprise-grade reinforcement learning framework for large-scale LLM and VLM post-training, forked from and co-ev… | 84 | 3169 | active |
| mjlab mjlab is a Python framework that combines Isaac Lab's manager-based API with MuJoCo Warp, a GPU-accelerated version of MuJoCo, for reinforc… | 82 | 3163 | active |
| heavyai/heavydb HeavyDB (formerly MapD/OmniSciDB) is an open-source SQL-based, relational, columnar database engine that uses CPUs and NVIDIA GPUs to query… | 88 | 3067 | active |
| luminal-ai/luminal Luminal is a high-performance general-purpose ML inference compiler written in Rust that lowers models to a minimal 15-op dataflow IR and c… | 77 | 3026 | active |
| deepseek-ai/DualPipe DualPipe is a Python library implementing a bidirectional pipeline parallelism algorithm that overlaps forward and backward computation wit… | 45 | 3017 | active |
| pytorch/ao TorchAO is a PyTorch-native library for model optimization through quantization and sparsity. It supports quantizing weights, gradients, op… | 87 | 3001 | active |
| Cloudpods Cloudpods is an open-source, Golang-based cloud-native unified multi/hybrid-cloud platform ('a cloud on clouds') that manages on-premise KV… | 77 | 2958 | active |
| huggingface/nanotron Nanotron is a minimalistic Python library from Hugging Face for pretraining large language models with 3D parallelism (data, tensor, and pi… | 63 | 2843 | active |
| antirez/h3.c A native C inference engine for the MiniMax H3 model on Apple Silicon, using Metal for GPU acceleration. It generates video (and audio) fro… | 53 | 2828 | active |
| Lepton AI LeptonAI is a Python library and `lep` CLI for operating NVIDIA DGX Cloud Lepton, a platform that unifies global GPU compute for AI develop… | 96 | 2824 | active |
| black-forest-labs/flux2 Official inference repository for Black Forest Labs' FLUX.2 family of open-weight image generation and editing models. It provides minimal … | 45 | 2720 | active |
| EMI-Group/evox EvoX is a distributed GPU-accelerated framework for evolutionary computation, compatible with PyTorch and built on JAX. It provides 50+ evo… | 97 | 2640 | active |
| google-coral/coralnpu Coral NPU is an open-source neural processing unit (NPU) hardware IP core from Google Research, built on the 32-bit RISC-V ISA with matrix,… | 72 | 2604 | active |
| data-infra/cube-studio CubeStudio is an open-source, cloud-native, all-in-one AI platform covering the full machine learning lifecycle (MLOps/MaaS/LLMOps), includ… | 75 | 2546 | active |