From ffd7d654857437848ba803a10a18b1ded5ab274a Mon Sep 17 00:00:00 2001 From: Anthony Shoumikhin Date: Wed, 21 Oct 2026 00:16:54 -0700 Subject: [PATCH] Docs: say which build pip installs from PyPI On Linux, the ExecuTorch package on PyPI becomes the CUDA build, the same choice PyTorch makes for its own PyPI package. macOS and Windows keep the CPU build. Say so in the README, in Getting Started and in the C++ guide, and point CPU-only Linux users at the cpu index. --- README.md | 5 +++++ docs/source/getting-started.md | 6 ++++++ docs/source/using-executorch-cpp.md | 3 ++- 3 files changed, 13 insertions(+), 1 deletion(-) diff --git a/README.md b/README.md index f920bc41169..b5ce3d55582 100644 --- a/README.md +++ b/README.md @@ -83,6 +83,11 @@ Install the latest stable Python package in a Python 3.11–3.15 environment: pip install executorch ``` +On Linux this is the CUDA build, matching the `torch` package on PyPI. On macOS +and Windows it is the CPU build. For the CPU-only Linux build or another CUDA +version, see +[Getting Started](https://docs.pytorch.org/executorch/main/getting-started.html). + Install a nightly built from `main` to use the newest features: ```bash diff --git a/docs/source/getting-started.md b/docs/source/getting-started.md index 15a3b92fd73..3ffde9e9f9a 100644 --- a/docs/source/getting-started.md +++ b/docs/source/getting-started.md @@ -44,6 +44,12 @@ across minor versions but not across major ones. There is no package for CUDA 12 so on CUDA 12 use the `cpu` variant or build from source. On CUDA 13.0 or 13.1 no lower variant exists, so use `cu132`. +Without `--index-url`, pip takes both packages from PyPI, which follows the same +layout for both. On Linux they are the CUDA build, for the CUDA version PyTorch +publishes there. On macOS and Windows they are the CPU build. The Linux packages +also work on a machine without a GPU, but they download the CUDA libraries. Use +the `cpu` variant to skip them. + To get a change that has landed on the `main` branch but is not in a release yet, use a nightly build. These are rebuilt every day. Put `nightly/` in front of the variant name, for example `nightly/cu132`, and add `--pre` to the diff --git a/docs/source/using-executorch-cpp.md b/docs/source/using-executorch-cpp.md index 221acdafd63..8b0c8df29b4 100644 --- a/docs/source/using-executorch-cpp.md +++ b/docs/source/using-executorch-cpp.md @@ -423,7 +423,8 @@ event tracer. ### Running on a GPU with the CUDA package -The CUDA build is a separate package. Releases cover CUDA 13.2 and 13.4, so pick the index +On Linux, the package on PyPI is the CUDA build, for the same CUDA version as the `torch` package +there. Releases also cover CUDA 13.2 and 13.4 on their own indexes, so you can pick the one matching the CUDA version you have (`cu132` or `cu134`). For CUDA 13.2: ```