[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-70790":3},{"id":4,"name":5,"fullName":6,"owner":5,"repo":5,"description":7,"homepage":8,"htmlUrl":9,"language":10,"languages":9,"totalLinesOfCode":9,"stars":11,"forks":12,"watchers":13,"openIssues":14,"contributorsCount":15,"subscribersCount":15,"size":15,"stars1d":16,"stars7d":17,"stars30d":18,"stars90d":15,"forks30d":15,"starsTrendScore":19,"compositeScore":20,"rankGlobal":9,"rankLanguage":9,"license":21,"archived":22,"fork":22,"defaultBranch":23,"hasWiki":24,"hasPages":22,"topics":25,"createdAt":9,"pushedAt":9,"updatedAt":43,"readmeContent":44,"aiSummary":45,"trendingCount":15,"starSnapshotCount":15,"syncStatus":46,"lastSyncTime":47,"discoverSource":48},70790,"cupy","cupy\u002Fcupy","NumPy & SciPy for GPU","https:\u002F\u002Fcupy.dev",null,"Python",10992,1036,130,549,0,6,19,54,18,44.05,"MIT License",false,"main",true,[26,27,28,5,29,30,31,32,33,34,35,36,37,38,39,40,41,42],"cublas","cuda","cudnn","curand","cusolver","cusparse","cusparselt","cutensor","gpu","nccl","numpy","nvrtc","nvtx","python","rocm","scipy","tensor","2026-06-12 02:02:43","\u003Cdiv align=\"center\">\u003Cimg src=\"https:\u002F\u002Fraw.githubusercontent.com\u002Fcupy\u002Fcupy\u002Fmain\u002Fdocs\u002Fimage\u002Fcupy_logo_1000px.png\" width=\"400\"\u002F>\u003C\u002Fdiv>\n\n# CuPy : NumPy & SciPy for GPU\n\n[![pypi](https:\u002F\u002Fimg.shields.io\u002Fpypi\u002Fv\u002Fcupy)](https:\u002F\u002Fpypi.python.org\u002Fpypi\u002Fcupy)\n[![Conda](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fconda--forge-cupy-blue)](https:\u002F\u002Fanaconda.org\u002Fconda-forge\u002Fcupy)\n[![GitHub license](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Flicense\u002Fcupy\u002Fcupy)](https:\u002F\u002Fgithub.com\u002Fcupy\u002Fcupy)\n[![Matrix](https:\u002F\u002Fimg.shields.io\u002Fmatrix\u002Fcupy_community:gitter.im?server_fqdn=matrix.org)](https:\u002F\u002Fgitter.im\u002Fcupy\u002Fcommunity)\n[![Twitter](https:\u002F\u002Fimg.shields.io\u002Ftwitter\u002Ffollow\u002FCuPy_Team?label=%40CuPy_Team)](https:\u002F\u002Ftwitter.com\u002FCuPy_Team)\n[![Medium](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FMedium-CuPy-teal)](https:\u002F\u002Fmedium.com\u002Fcupy-team)\n\n[**Website**](https:\u002F\u002Fcupy.dev\u002F)\n| [**Install**](https:\u002F\u002Fdocs.cupy.dev\u002Fen\u002Fstable\u002Finstall.html)\n| [**Tutorial**](https:\u002F\u002Fdocs.cupy.dev\u002Fen\u002Fstable\u002Fuser_guide\u002Fbasic.html)\n| [**Examples**](https:\u002F\u002Fgithub.com\u002Fcupy\u002Fcupy\u002Ftree\u002Fmain\u002Fexamples)\n| [**Documentation**](https:\u002F\u002Fdocs.cupy.dev\u002Fen\u002Fstable\u002F)\n| [**API Reference**](https:\u002F\u002Fdocs.cupy.dev\u002Fen\u002Fstable\u002Freference\u002F)\n| [**Forum**](https:\u002F\u002Fgroups.google.com\u002Fforum\u002F#!forum\u002Fcupy)\n\nCuPy is a NumPy\u002FSciPy-compatible array library for GPU-accelerated computing with Python.\nCuPy acts as a [drop-in replacement](https:\u002F\u002Fdocs.cupy.dev\u002Fen\u002Fstable\u002Freference\u002Fcomparison.html) to run existing NumPy\u002FSciPy code on NVIDIA CUDA or AMD ROCm platforms.\n\n```py\n>>> import cupy as cp\n>>> x = cp.arange(6).reshape(2, 3).astype('f')\n>>> x\narray([[ 0.,  1.,  2.],\n       [ 3.,  4.,  5.]], dtype=float32)\n>>> x.sum(axis=1)\narray([  3.,  12.], dtype=float32)\n```\n\nCuPy also provides access to low-level CUDA features.\nYou can pass `ndarray` to existing CUDA C\u002FC++ programs via [RawKernels](https:\u002F\u002Fdocs.cupy.dev\u002Fen\u002Fstable\u002Fuser_guide\u002Fkernel.html#raw-kernels), use [Streams](https:\u002F\u002Fdocs.cupy.dev\u002Fen\u002Fstable\u002Freference\u002Fcuda.html) for performance, or even call [CUDA Runtime APIs](https:\u002F\u002Fdocs.cupy.dev\u002Fen\u002Fstable\u002Freference\u002Fcuda.html#runtime-api) directly.\n\n## Installation\n\n### Pip\n\nBinary packages (wheels) are available for Linux and Windows on [PyPI](https:\u002F\u002Fpypi.org\u002Forg\u002Fcupy\u002F).\nChoose the right package for your platform.\n\n| Platform                                                                                                     | Architecture      | Command                     |\n|--------------------------------------------------------------------------------------------------------------| ----------------- |-----------------------------|\n| CUDA 12.x                                                                                                    | x86_64 \u002F aarch64  | `pip install cupy-cuda12x`  |\n| CUDA 13.x                                                                                                    | x86_64 \u002F aarch64  | `pip install cupy-cuda13x`  |\n| ROCm 7.0 (*[experimental](https:\u002F\u002Fdocs.cupy.dev\u002Fen\u002Flatest\u002Finstall.html#using-cupy-on-amd-gpu-experimental)*) | x86_64            | `pip install cupy-rocm-7-0` |\n\n> [!NOTE]\\\n> To install pre-releases, append `--pre -U -f https:\u002F\u002Fpip.cupy.dev\u002Fpre` (e.g., `pip install cupy-cuda12x --pre -U -f https:\u002F\u002Fpip.cupy.dev\u002Fpre`).\n\n### Conda\n\nBinary packages are also available for Linux and Windows on [Conda-Forge](https:\u002F\u002Fanaconda.org\u002Fconda-forge\u002Fcupy).\n\n| Platform              | Architecture                | Command                                                       |\n| --------------------- | --------------------------- | ------------------------------------------------------------- |\n| CUDA                  | x86_64 \u002F aarch64 \u002F ppc64le  | `conda install -c conda-forge cupy`                           |\n\nIf you need a slim installation (without also getting CUDA dependencies installed), you can do `conda install -c conda-forge cupy-core`.\n\nIf you need to use a particular CUDA version (say 12.0), you can use the `cuda-version` metapackage to select the version, e.g. `conda install -c conda-forge cupy cuda-version=12.0`.\n\n> [!NOTE]\\\n> If you encounter any problem with CuPy installed from `conda-forge`, please feel free to report to [cupy-feedstock](https:\u002F\u002Fgithub.com\u002Fconda-forge\u002Fcupy-feedstock\u002Fissues), and we will help investigate if it is just a packaging issue in `conda-forge`'s recipe or a real issue in CuPy.\n\n### Docker\n\nUse [NVIDIA Container Toolkit](https:\u002F\u002Fdocs.nvidia.com\u002Fdatacenter\u002Fcloud-native\u002Fcontainer-toolkit\u002Foverview.html) to run [CuPy container images](https:\u002F\u002Fhub.docker.com\u002Fr\u002Fcupy\u002Fcupy).\n\n```\n$ docker run --gpus all -it cupy\u002Fcupy\n```\n\n## Resources\n\n- [Installation Guide](https:\u002F\u002Fdocs.cupy.dev\u002Fen\u002Fstable\u002Finstall.html) - instructions on building from source\n- [Release Notes](https:\u002F\u002Fgithub.com\u002Fcupy\u002Fcupy\u002Freleases)\n- [Projects using CuPy](https:\u002F\u002Fgithub.com\u002Fcupy\u002Fcupy\u002Fwiki\u002FProjects-using-CuPy)\n- [Contribution Guide](https:\u002F\u002Fdocs.cupy.dev\u002Fen\u002Fstable\u002Fcontribution.html)\n- [GPU Acceleration in Python using CuPy and Numba (GTC November 2021 Technical Session)](https:\u002F\u002Fwww.nvidia.com\u002Fen-us\u002Fon-demand\u002Fsession\u002Fgtcfall21-a31149\u002F)\n- [GPU-Acceleration of Signal Processing Workflows using CuPy and cuSignal[^1] (ICASSP'21 Tutorial)](https:\u002F\u002Fgithub.com\u002Fawthomp\u002Fcusignal-icassp-tutorial)\n\n[^1]: cuSignal is now part of CuPy starting v13.0.0.\n\n## License\n\nMIT License (see `LICENSE` file).\n\nCuPy is designed based on NumPy's API and SciPy's API (see `docs\u002Fsource\u002Flicense.rst` file).\n\nCuPy is being developed and maintained by [Preferred Networks](https:\u002F\u002Fwww.preferred.jp\u002Fen\u002F) and [community contributors](https:\u002F\u002Fgithub.com\u002Fcupy\u002Fcupy\u002Fgraphs\u002Fcontributors).\n\n## Reference\n\nRyosuke Okuta, Yuya Unno, Daisuke Nishino, Shohei Hido and Crissman Loomis.\n**CuPy: A NumPy-Compatible Library for NVIDIA GPU Calculations.**\n*Proceedings of Workshop on Machine Learning Systems (LearningSys) in The Thirty-first Annual Conference on Neural Information Processing Systems (NIPS)*, (2017).\n[[PDF](http:\u002F\u002Flearningsys.org\u002Fnips17\u002Fassets\u002Fpapers\u002Fpaper_16.pdf)]\n\n```bibtex\n@inproceedings{cupy_learningsys2017,\n  author       = \"Okuta, Ryosuke and Unno, Yuya and Nishino, Daisuke and Hido, Shohei and Loomis, Crissman\",\n  title        = \"CuPy: A NumPy-Compatible Library for NVIDIA GPU Calculations\",\n  booktitle    = \"Proceedings of Workshop on Machine Learning Systems (LearningSys) in The Thirty-first Annual Conference on Neural Information Processing Systems (NIPS)\",\n  year         = \"2017\",\n  url          = \"http:\u002F\u002Flearningsys.org\u002Fnips17\u002Fassets\u002Fpapers\u002Fpaper_16.pdf\"\n}\n```\n","CuPy 是一个兼容 NumPy 和 SciPy 的 GPU 加速数组库，专为 Python 编程语言设计。它能够无缝替代现有的 NumPy\u002FSciPy 代码，在 NVIDIA CUDA 或 AMD ROCm 平台上运行，从而显著提高计算密集型任务的执行效率。除了提供与 NumPy 相似的高级功能外，CuPy 还支持直接访问低级别的 CUDA 特性，如通过 RawKernels 传递 `ndarray` 给 CUDA C\u002FC++ 程序、使用 Streams 优化性能以及调用 CUDA Runtime API。这使得 CuPy 不仅适用于科学计算和数据分析领域内的常规应用，还非常适合需要高性能并行处理能力的深度学习、图像处理等场景。",2,"2026-06-11 03:34:13","high_star"]