{"full_name":"pytorch/pytorch","owner":"pytorch","name":"pytorch","url":"https://github.com/pytorch/pytorch","description":"Tensors and Dynamic neural networks in Python with strong GPU acceleration","homepage":"https://pytorch.org","topics":[],"language":"Python","license":"NOASSERTION","avatar":"https://avatars.githubusercontent.com/u/21003710?v=4","created_at":"2016-08-13T05:26:41Z","pushed_at":"2026-11-10T09:28:42Z","metrics":{"stars":105307,"forks":30618,"watchers":1792,"issues":62391,"prs":140888,"releases":71,"issues_open":null,"issues_closed":null,"prs_open":null,"prs_closed":null,"prs_merged":null,"discussions":0},"growth":{"day":{"stars":30,"forks":-455,"base_day":"2026-11-09"},"week":{"stars":222,"forks":-699,"base_day":"2026-11-03"},"month":{"stars":1530,"forks":-857,"base_day":"2026-10-11"}},"history":{"dates":["2026-09-06","2026-09-07","2026-09-08","2026-09-09","2026-09-10","2026-09-11","2026-09-12","2026-09-13","2026-09-14","2026-09-15","2026-09-16","2026-09-17","2026-09-18","2026-09-19","2026-09-20","2026-09-21","2026-09-22","2026-09-23","2026-09-24","2026-09-25","2026-09-26","2026-09-27","2026-09-28","2026-09-29","2026-09-30","2026-10-01","2026-10-02","2026-10-03","2026-10-04","2026-10-05","2026-10-06","2026-10-07","2026-10-08","2026-10-09","2026-10-10","2026-10-11","2026-10-12","2026-10-13","2026-10-14","2026-10-15","2026-10-16","2026-10-17","2026-10-18","2026-10-19","2026-10-20","2026-10-21","2026-10-22","2026-10-23","2026-10-24","2026-10-25","2026-10-26","2026-10-27","2026-10-28","2026-10-29","2026-10-30","2026-10-31","2026-11-01","2026-11-02","2026-11-03","2026-11-04","2026-11-09","2026-11-10"],"stars":[102577,102589,102599,102623,102635,102651,102658,102707,102727,102740,102760,102771,102845,102863,102874,102910,102930,102953,103006,103026,103057,103072,103084,103107,103153,103177,103200,103250,103310,103365,103462,103523,103601,103621,103667,103777,103821,103847,103901,103959,104094,104730,104830,104670,104699,104796,104824,104915,104961,104963,104978,105006,105039,105092,105119,105146,104999,105023,105085,105104,105277,105307],"forks":[28969,28980,28992,29013,29024,29024,29031,29073,29090,29100,29108,29115,29152,29165,29165,29181,29196,29207,29224,29286,29488,29582,29730,29852,30003,30021,30080,30198,30408,30552,30812,31001,31030,31104,31200,31475,31512,31602,31612,31747,31921,32192,32311,32490,32530,32706,32717,32912,32672,32450,32511,32469,32335,31788,31808,31844,31855,31884,31317,31310,31073,30618],"watchers":[1805,1804,1804,1804,1804,1804,1805,1807,1806,1806,1805,1805,1807,1807,1800,1801,1802,1802,1802,1803,1802,1803,1803,1802,1803,1804,1802,1802,1802,1802,1803,1801,1800,1800,1800,1798,1797,1797,1798,1805,1805,1804,1813,1813,1813,1813,1813,1814,1810,1804,1804,1803,1798,1795,1795,1794,1792,1793,1793,1792,1792,1792],"issues_open":[null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null],"prs_open":[null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null]},"boards":{"total":133},"same_owner":[{"full_name":"pytorch/torchtitan","name":"torchtitan","boards":{}},{"full_name":"pytorch/tutorials","name":"tutorials","boards":{}},{"full_name":"pytorch/vision","name":"vision","boards":{}}],"issues":[{"number":77764,"title":"General MPS op coverage tracking issue","comments":1797,"created_at":"2022-05-18T18:12:47Z"},{"number":93531,"title":"TorchInductor CPU Performance Dashboard","comments":1296,"created_at":"2022-10-13T02:29:48Z"},{"number":141287,"title":"MPS operator coverage tracking issue (2.6+ version)","comments":268,"created_at":"2024-11-21T21:15:46Z"},{"number":191009,"title":"DISABLED test_index_propagation_to_dtype_inf_cpu_halide (__main__.HalideCpuTests)","comments":190,"created_at":"2026-07-24T13:06:20Z"},{"number":47702,"title":"GPU acceleration for Apple's M1 chip?","comments":183,"created_at":"2020-11-10T22:05:54Z"},{"number":13246,"title":"DataLoader num_workers > 0 causes CPU memory from parent process to be replicated in all worker processes","comments":165,"created_at":"2018-10-29T13:23:59Z"},{"number":160230,"title":"Offer official Pytorch Vulkan backend on pytorch.org","comments":132,"created_at":"2025-08-08T23:19:51Z"},{"number":198596,"title":"DISABLED test_rerun_then_pass (test.junit_xml_testdata.pytest_suite.TestJunitOutcomes)","comments":121,"created_at":"2026-09-25T06:57:58Z"}],"discussions":[],"latest_release":{"name":"PyTorch 2.15.0 Release","tag":"v2.15.0","published_at":"2026-10-28T17:09:49Z","body":"# PyTorch 2.15.0 Release Notes\n\n- [Highlights](#highlights)\n- [Backwards Incompatible Changes](#backwards-incompatible-changes)\n- [Deprecations](#deprecations)\n- [New Features](#new-features)\n- [Improvements](#improvements)\n- [Bug fixes](#bug-fixes)\n- [Performance](#performance)\n- [Documentation](#documentation)\n- [Security](#security)\n- [Developers](#developers)\n\n# Highlights\n\n<table>\n  <tr><td><strong>Rewritten NCCL backend becomes the default</strong>, with a flight recorder that dumps on hangs and <code>TORCH_DIST_USE_NCCL2=0</code> or <code>nccl-legacy</code> as rollback paths</td></tr>\n  <tr><td><strong>One-sided transports for <code>torch.distributed</code></strong>, with a NIXL backend and CUDA-graph-capturable transfers</td></tr>\n  <tr><td><strong>FSDP2 and pipelining gain finer memory and communication control</strong>, with fp32 for precision-sensitive parameters, CUDA-graph-captured gradient accumulation, and gradient reduction overlapped with compute</td></tr>\n  <tr><td><strong>Faster 4-bit (NVFP4) LLM decoding on NVIDIA Blackwell GPUs</strong> through fewer, better-overlapped kernel launches in NVGEMM</td></tr>\n  <tr><td><strong>Apple Silicon gains FP8 and faster decode</strong>, with <code>float8_e4m3fn</code> and <code>_scaled_mm</code>, a new <code>grouped_mm</code> with a decode fast path (1.32–1.34x tokens/s on MoE decode), and public <code>torch.mps.Stream</code> APIs</td></tr>\n  <tr><td><strong>Broader platform support</strong> with FlyDSL kernels on AMD gfx950, Tensor Data Mover support for gfx1250 (MI450), ROCm 10.0 wheels, simpler SYCL kernel authoring and multi-GPU profiling on Intel GPUs, CUDA 13.2 as the default wheel with CUDA 13.4 alongside it, and <code>torch.compile</code> on Python 3.15</td></tr>\n</table>\n\nFor more details about these highlighted features, you can look at the release blogpost. Below are the full release notes for this release.\n\n# Backwards Incompatible Changes\n\n## Python Frontend\n\n### `import torch` can no longer be re-executed in the same process, including via `importlib.reload(torch)` (#194654)\n\n`importlib.reload(torch)` and `del sys.modules[\"torch\"]` + re-import now raise, as does retrying `import torch` after it failed. Re-running the module body was already unsafe in exactly the same way as the retry, so the new check does not distinguish the two. Normal use is unaffected: a successful import leaves `torch` in `sys.modules` so the body never re-runs, and re-entrant imports from a backend extension during autoload still short-circuit on `sys.modules` without reaching the check.\n\nVersion 2.14:\n```python\nimport importlib\nimport torch\n\nimportlib.reload(torch)\n# RuntimeError: Only a single TORCH_LIBRARY can be used to register the namespace triton; ...\n# (re-importing after a failed `import torch` failed with similarly misleading errors or could segfault)\n```\n\nVersion 2.15:\n```python\nimport importlib\nimport torch\n\nimportlib.reload(torch)\n# ImportError: `torch` can only be initialized once per process, so this module cannot\n# be imported again: neither via `importlib.reload(torch)`, nor by re-importing after\n# a failed `import torch`.\n```\n\n### Nested tensors are no longer allowlisted for `torch.load(..., weights_only=True)` (#194380)\n\nNested jagged and strided nested tensors are no longer on the default `weights_only` allowlist. Loading a checkpoint that contains one with `weights_only=True` (the default) now fails with `Unsupported global: GLOBAL torch.nested._internal.nested_tensor._rebuild_njt was not an allowed global by default` (or `GLOBAL torch._utils._rebuild_nested_tensor` for strided nested tensors). To load checkpoints containing nested tensors, allow them explicitly with `torch.serialization.safe_globals` (see below), or use `weights_only=False` for trusted files.\n\nVersion 2.14:\n```python\nloaded = torch.load(\"njt.pt\", weights_only=True)  # works\n```\n\nVersion 2.15:\n```python\nfrom torch.nested._internal.nested_tensor import _rebuild_njt, NestedTensor\nwith torch.seriali"},"readme":true,"readme_zh":false,"news":[{"title":"PyTorch 1.0 is out","url":"https://github.com/pytorch/pytorch/releases/tag/v1.0.0","hn":"https://news.ycombinator.com/item?id=18630721","source":"Hacker News","lang":"en","date":"2018-12-07","points":470,"comments":70},{"title":"PyTorch 1.8, with AMD ROCm support","url":"https://github.com/pytorch/pytorch/releases/tag/v1.8.0","hn":"https://news.ycombinator.com/item?id=26353184","source":"Hacker News","lang":"en","date":"2021-03-05","points":313,"comments":138},{"title":"PyTorch v0.2.0 released","url":"https://github.com/pytorch/pytorch/releases/tag/v0.2.0","hn":"https://news.ycombinator.com/item?id=14946585","source":"Hacker News","lang":"en","date":"2017-08-07","points":150,"comments":14},{"title":"PyTorch 1.1","url":"https://github.com/pytorch/pytorch/releases/tag/v1.1.0","hn":"https://news.ycombinator.com/item?id=19793425","source":"Hacker News","lang":"en","date":"2019-05-01","points":43,"comments":2},{"title":"PyTorch 1.4: Custom mobile builds, distributed model parallel, Java bindings","url":"https://github.com/pytorch/pytorch/releases/tag/v1.4.0","hn":"https://news.ycombinator.com/item?id=22061894","source":"Hacker News","lang":"en","date":"2020-01-16","points":39,"comments":4},{"title":"Pytorch 0.4.0 is out","url":"https://github.com/pytorch/pytorch/releases/tag/v0.4.0","hn":"https://news.ycombinator.com/item?id=16916687","source":"Hacker News","lang":"en","date":"2018-04-24","points":36,"comments":7},{"title":"PyTorch 1.7 Released with CUDA 11, New APIs for FFTs, Win Distributed Training","url":"https://github.com/pytorch/pytorch/releases/tag/v1.7.0","hn":"https://news.ycombinator.com/item?id=24909133","source":"Hacker News","lang":"en","date":"2020-10-27","points":29,"comments":1},{"title":"PyTorch 1.8 Release, including Compiler and Distributed Training updates","url":"https://github.com/pytorch/pytorch/releases/tag/v1.8.0","hn":"https://news.ycombinator.com/item?id=26348170","source":"Hacker News","lang":"en","date":"2021-03-04","points":6,"comments":1},{"title":"Google Open-Sources TorchTPU, Putting Native PyTorch on Its TPUs","url":"https://www.unite.ai/google-open-sources-torchtpu-putting-native-pytorch-on-its-tpus/","source":"unite","lang":"en","date":"2026-10-20"},{"title":"AMD and Intel Consolidate PyTorch CPU Optimizations in oneDNN","url":"https://xenospectrum.com/en/amd-intel-pytorch-onednn-integration/","source":"xenospectrum","lang":"en","date":"2026-10-20"},{"title":"Google open-sourced TorchTPU, so PyTorch on TPUs just got a lot easier","url":"https://www.helpnetsecurity.com/2026/10/21/torchtpu-pytorch-on-tpus/","source":"Help Net Security","lang":"en","date":"2026-10-20"},{"title":"PyTorch Tutorial for Deep Learning","url":"https://blog.jetbrains.com/pycharm/2026/07/pytorch-tutorial-for-deep-learning/","source":"blog.jetbrains.com","lang":"en","date":"2026-07-29"},{"title":"PyTorch vs. TensorFlow: Choosing the Right Framework in 2026","url":"https://blog.jetbrains.com/pycharm/2026/05/pytorch-vs-tensorflow-choosing-framework-2026/","source":"blog.jetbrains.com","lang":"en","date":"2026-05-03"},{"title":"PyTorch Foundation Welcomes Ray to Deliver a Unified Open Source AI Compute Stack","url":"https://www.prnewswire.com/news-releases/pytorch-foundation-welcomes-ray-to-deliver-a-unified-open-source-ai-compute-stack-302591184.html","source":"irblog.prnewswire.com","lang":"en","date":"2025-10-22"},{"title":"PyTorch 2.0 brings new fire to open-source machine learning","url":"https://venturebeat.com/technology/pytorch-2-0-brings-new-fire-to-open-source-machine-learning","source":"VentureBeat","lang":"en","date":"2023-03-15"},{"title":"PyTorch 2.0 release accelerates open-source machine learning","url":"https://venturebeat.com/programming-development/pytorch-2-0-release-accelerates-open-source-machine-learning","source":"VentureBeat","lang":"en","date":"2022-12-02"},{"title":"What is PyTorch? Python machine learning on GPUs","url":"https://www.infoworld.com/article/2335194/what-is-pytorch-python-machine-learning-on-gpus.html","source":"InfoWorld","lang":"en","date":"2022-05-06"},{"title":"PyTorch 1.9 has arrived: Here's what you need to know","url":"https://www.zdnet.com/article/pytorch-1-9-has-arrived-heres-what-you-need-to-know/","source":"ZDNet","lang":"en","date":"2021-06-16"},{"title":"Microsoft adds enterprise support for PyTorch AI on Azure","url":"https://www.zdnet.com/article/microsoft-adds-enterprise-support-for-pytorch-ai-on-azure/","source":"ZDNet","lang":"en","date":"2021-05-25"},{"title":"TorchTPU is now open source: Run native PyTorch on TPUs","url":"https://news.google.com/rss/articles/CBMirgFBVV95cUxPamRXcjBrRXo2V1plWFFvajhjcUFEa0N4aVZPN3dvVGlPTzZwbF9oUlFWNC1vY1EzdTc1M3pVb3QwZERTekNoU3ZEekd0WlU5czBFUF9qc01aOGhkanpTZGFxeVdKTDM5Tks1Q19uZUtYQ1RUQzFCRjYyTGMtbmg4RkQwUzUzYmpaMllDRERqU21aVkg2cU8wYkI4eGZWWEpOYzdkZmtxTEw5enI5LVE?oc=5","source":"Google Cloud","lang":"en","date":"2026-10-20"},{"title":"Mini Shai-Hulud: A Cross-Ecosystem Supply Chain Attack on PyTorch Lightning & Intercom Client","url":"https://news.google.com/rss/articles/CBMizgFBVV95cUxPdmRPWGF6eXZCWDkwTlk5aUZNeXI3QkN6a21NVUZxOGQtdDJnNnB0VXFULUVsX1VKdGJzYWVmY3JlbWljbjNYZm9jY0NFY3F6TENwa3UzLWY0elNqdkZyS19BUmJISjFrWjRBNTdrNG5tUk5QdEJ6dW1UT2RkZlBhMWVKR09FMVhQT1NQMjdqMXRKbGxlbWUyVTljbEw3R3plUkNHRlVWM1FHcU01NXVEa0NmbEo5cXU2R2dkVFZXYlNSRG9ZalE2MjFiUzc4QQ?oc=5","source":"Resecurity","lang":"en","date":"2026-05-09"},{"title":"PyTorch Lightning and Intercom-client Hit in Supply Chain Attacks to Steal Credentials","url":"https://news.google.com/rss/articles/CBMiggFBVV95cUxPWlRieFFjOTZTa0U0RmNSZFQyWndWcmhTTHdTN3hqdF9WY1B2MGdWbGlrMDhaWDRKajYzVzExa1QtUFo4MXRQMG5wS2Jicmp4dFByRzRyUV9KYVFyYWZ3RjA2U2piX2ppMGFjNFJqSHJfeGVmbUFuNTlpdzFuaGxpOWVR?oc=5","source":"The Hacker News","lang":"en","date":"2026-04-30"},{"title":"Meet OpenMythos: An Open-Source PyTorch Reconstruction of Claude Mythos Where 770M Parameters Match a 1.3B Transformer","url":"https://news.google.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?oc=5","source":"MarkTechPost","lang":"en","date":"2026-04-19"},{"title":"Rust 机器学习，一定要知道的这些库，可以替代 Python 库了","url":"https://news.google.com/rss/articles/CBMiU0FVX3lxTFBfcnB3WXItQURXd0pxVmx2OHg0a0VNWWFkQ2dKZ2tqNlhpMFlJTjFCT195V2l5Vm1kcmREN3FHc3ZYVE5xWElmRy10WE1oWUxsUUlz?oc=5","source":"51CTO","lang":"zh","date":"2024-04-26"},{"title":"GitHub 7.5k star量，各种视觉Transformer的PyTorch实现合集整理好了","url":"https://news.google.com/rss/articles/CBMiSEFVX3lxTE1ISnBBUW1LeXFvSTJWNjhSbkFxTU9KZkYyU1RMZlU4dVRvUC1tLWcxUGpCM3lkZnNhMWEzT3V6WkViWEd2YzE0RA?oc=5","source":"智源社区","lang":"zh","date":"2021-12-31"},{"title":"可以在手机里运行的Detectron2来了：FB官方出品","url":"https://news.google.com/rss/articles/CBMiU0FVX3lxTFBNdTgyMFZvWFRqNFVJS3B2ek5HU1BkSFpnYWluRzIzcGVrUHN5QWNGc2syV2dkZDRnRXQ1LVFGM3ZZaXlYeWxsRm10aXh4b3o1YXdr?oc=5","source":"量子位","lang":"zh","date":"2021-03-06"},{"title":"GitHub 1400星，天才黑客开源深度学习框架tinygrad","url":"https://news.google.com/rss/articles/CBMiTEFVX3lxTFBabXZRampxbHZjamc0TS1ZenB4VVRjWEhlRGhuOUJLeTV6dm54Vjg2UXB5NWZtRHJLZFl4VThaRVkzdFNKeGdYY2tBNWs?oc=5","source":"36Kr","lang":"zh","date":"2020-11-26"},{"title":"GitHub发布年度机器学习榜：TensorFlow最火，PyTorch失踪，Julia第四","url":"https://news.google.com/rss/articles/CBMiUEFVX3lxTE1QWXRBeHRVNEI5My15bTFUdUEwbVlFblg0ek4tbGFCZ3FEOWJieEVsUXQ4SURxa19Vem1UN0tOZVRkc1hoNEY5Mk9rVmhyeXpr?oc=5","source":"搜狐网","lang":"zh","date":"2019-01-26"},{"title":"迎来 PyTorch，告别 Theano，2017 深度学习框架发展大盘点","url":"https://news.google.com/rss/articles/CBMib0FVX3lxTE5oRC1menZVeVl3ejh0b2V6LXQ4QVJPZkJiZWJJcDJrYjNvVE16RGRWcUxyYzk2aHNXTzRlZ1U5TVM0NzBUak9JOU1aZnZERUdLLVNOQVl6dFg4WnhhbVJNUnl5cVdzZGZ3RjFIY3p0RQ?oc=5","source":"雷峰网","lang":"zh","date":"2017-12-31"}],"stale":false,"snapshot_at":"2026-11-10T09:35:25Z","history_start":"2026-09-06"}