Publish CUDA wheels so a GPU user does not build from source

[shoumikhin]

Right now pip install executorch always gives you a CPU-only wheel.
The CUDA delegate is built and tested from source, but no published wheel contains it, so anyone
who wants GPU support has to clone the repository and build.

This turns on the GPU wheel rows so a plain install can bring the delegate with it:

pip install executorch                              # CPU, unchanged
pip install executorch --index-url <cuda index>     # CUDA delegate included

Two things make those wheels correct rather than merely present

The GPU generations to compile for. Without being told, the build compiles device code only
for the GPU in the build machine. The wheel then installs fine on every machine the row claims
and fails when a model actually runs on a different generation, which looks like a model bug
rather than a packaging one. Each row now names its architectures explicitly:

x86_64  CUDA 13    A100, H100, B200, RTX 50 series
aarch64 CUDA 13    Grace-Hopper, GB200, and JetPack 7 devices
        CUDA 12.6  Jetson Orin

The CUDA libraries the wheel needs. The delegate links the CUDA runtime but does not bundle
it, so the wheel now declares those libraries as dependencies and one install brings what it
needs. The build variant is read from the release environment rather than guessed from whether a
CUDA compiler happens to be installed, because a CPU wheel built on a machine that has one must
not declare them.

Tested

Built a GPU wheel for each supported CUDA version on Linux x86_64 and on two Jetson devices,
then confirmed the built wheel carries device code for every architecture its row claims and
declares the matching CUDA packages. A CPU wheel is unchanged: no architectures named, no CUDA
dependencies declared.

[shoumikhin]

Stack from ghstack (oldest at bottom):

• Check the shipped device code covers the GPUs a wheel claims #21567
• -> Publish CUDA wheels so a GPU user does not build from source #21569
• Ship the CUDA delegate as its own library #21531
• Ship the XNNPACK delegate as its own library #21526
• Ship the merged CPU kernels as their own library #21524
• Ship the thread pool as its own library so a process has one pool #21522
• Ship a linkable shared runtime and make the Python bindings use it #21514

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/21569

• 📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

❌ 16 New Failures, 340 Pending, 40 Unrelated Failures, 31 Unclassified Failures

As of commit fc8b4e3 with merge base ad3a71f ():

NEW FAILURES - The following jobs have failed:

• nightly / link-check / lint-urls (gh)
• nightly / link-check / lint-xrefs (gh)
• pull / test-arm-backend-public-api-backward-compatibility / linux-job (gh)
  ##[error]API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID BB40:36EB96:22373D:72DF79:6A72698F and timestamp 2026-08-04 22:37:03 UTC. For more on scraping GitHub and how it may affect your rights, please review our Terms of Service (https://docs.github.com/en/site-policy/github-terms/github-terms-of-service) - https://docs.github.com/en/rest/using-the-rest-api/getting-started-with-the-rest-api#rate-limiting
• pull / test-multimodal-linux (gemma3-4b) / linux-job (gh)
  RuntimeError: Command docker exec -t c568a3c984a26103fa65c31f0f7bb7ecb4eaa76897fe7caa4c1bc87ac336821c /exec failed with exit code 1
• pull / unittest / linux / linux-job (gh)
  ##[error]API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID E6F4:2426C5:8E624A:1DCA704:6A72698C and timestamp 2026-08-04 22:37:01 UTC. For more on scraping GitHub and how it may affect your rights, please review our Terms of Service (https://docs.github.com/en/site-policy/github-terms/github-terms-of-service) - https://docs.github.com/en/rest/using-the-rest-api/getting-started-with-the-rest-api#rate-limiting
• pull / unittest-buck / linux / linux-job (gh)
  ##[error]API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID DFC8:7F1D:90F75E:1E7F749:6A7269AD and timestamp 2026-08-04 22:37:33 UTC. For more on scraping GitHub and how it may affect your rights, please review our Terms of Service (https://docs.github.com/en/site-policy/github-terms/github-terms-of-service) - https://docs.github.com/en/rest/using-the-rest-api/getting-started-with-the-rest-api#rate-limiting
• pull / unittest-editable / linux / linux-job (gh)
  ##[error]API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID BD40:26DEAF:212E95:6F3FFC:6A72698F and timestamp 2026-08-04 22:37:03 UTC. For more on scraping GitHub and how it may affect your rights, please review our Terms of Service (https://docs.github.com/en/site-policy/github-terms/github-terms-of-service) - https://docs.github.com/en/rest/using-the-rest-api/getting-started-with-the-rest-api#rate-limiting
• trunk / test-arm-backend-ethos-u (test_ootb_tests_ethos_u) / linux-job (gh)
  ##[error]API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID 9470:36EB96:205E79:6CA20C:6A726932 and timestamp 2026-08-04 22:35:30 UTC. For more on scraping GitHub and how it may affect your rights, please review our Terms of Service (https://docs.github.com/en/site-policy/github-terms/github-terms-of-service) - https://docs.github.com/en/rest/using-the-rest-api/getting-started-with-the-rest-api#rate-limiting
• trunk / test-arm-backend-ethos-u (test_pytest_ops_ethos_u55) / linux-job (gh)
  ##[error]API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID E1B0:349A77:1EE6465:670CCBA:6A726931 and timestamp 2026-08-04 22:35:30 UTC. For more on scraping GitHub and how it may affect your rights, please review our Terms of Service (https://docs.github.com/en/site-policy/github-terms/github-terms-of-service) - https://docs.github.com/en/rest/using-the-rest-api/getting-started-with-the-rest-api#rate-limiting
• trunk / test-arm-backend-ethos-u (test_run_ethos_u85) / linux-job (gh)
  ##[error]API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID CD8A:2BD68D:8A8168:1D27FF6:6A726935 and timestamp 2026-08-04 22:35:33 UTC. For more on scraping GitHub and how it may affect your rights, please review our Terms of Service (https://docs.github.com/en/site-policy/github-terms/github-terms-of-service) - https://docs.github.com/en/rest/using-the-rest-api/getting-started-with-the-rest-api#rate-limiting
• trunk / test-arm-backend-vkml (test_ootb_tests_vgf) / linux-job (gh)
  ##[error]An error occurred trying to start process '/usr/bin/bash' with working directory '/home/ec2-user/actions-runner/_work/executorch/executorch/pytorch/executorch'. No such file or directory
• trunk / test-arm-backend-vkml (test_smaller_stories_llama_vkml) / linux-job (gh)
  ##[error]API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID A6FA:22DC42:23D9FA:7566BC:6A726968 and timestamp 2026-08-04 22:36:24 UTC. For more on scraping GitHub and how it may affect your rights, please review our Terms of Service (https://docs.github.com/en/site-policy/github-terms/github-terms-of-service) - https://docs.github.com/en/rest/using-the-rest-api/getting-started-with-the-rest-api#rate-limiting
• trunk / test-cortex-m-e2e / run (mv3, cortex-m7) / mv3-cortex-m7 (gh)
  ##[error]API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID 8DAC:33EC7A:8C3DE7:1D8AAF1:6A7269A9 and timestamp 2026-08-04 22:37:30 UTC. For more on scraping GitHub and how it may affect your rights, please review our Terms of Service (https://docs.github.com/en/site-policy/github-terms/github-terms-of-service) - https://docs.github.com/en/rest/using-the-rest-api/getting-started-with-the-rest-api#rate-limiting
• trunk / test-huggingface-transformers-xnnpack (qwen3-0.6b|xnnpack|--quantize) / linux-job (gh)
  ##[error]An error occurred trying to start process '/usr/bin/bash' with working directory '/home/ec2-user/actions-runner/_work/executorch/executorch/pytorch/executorch'. No such file or directory
• trunk / unittest-release / linux / linux-job (gh)
  ##[error]API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID 8298:2B17BF:2063FAC:6BFAA6E:6A726974 and timestamp 2026-08-04 22:36:36 UTC. For more on scraping GitHub and how it may affect your rights, please review our Terms of Service (https://docs.github.com/en/site-policy/github-terms/github-terms-of-service) - https://docs.github.com/en/rest/using-the-rest-api/getting-started-with-the-rest-api#rate-limiting
• trunk / unittest-release / windows / windows-job (gh)
  Error: Input required and not supplied: path

UNCLASSIFIED FAILURES - DrCI could not classify the following jobs because the workflow did not run on the merge base. The failures may be pre-existing on trunk or introduced by this PR:

• Build Aarch64 Linux Wheels / pytorch/executorch / build-wheel-py3_10-cuda-aarch6412_6-aarch64 (gh) (this job did not run on the merge base, so DrCI cannot tell whether the failure is pre-existing)
• Build Aarch64 Linux Wheels / pytorch/executorch / build-wheel-py3_10-cuda-aarch6413_0-aarch64 (gh) (this job did not run on the merge base, so DrCI cannot tell whether the failure is pre-existing)
• Build Aarch64 Linux Wheels / pytorch/executorch / build-wheel-py3_10-cuda-aarch6413_2-aarch64 (gh) (this job did not run on the merge base, so DrCI cannot tell whether the failure is pre-existing)
• Build Aarch64 Linux Wheels / pytorch/executorch / build-wheel-py3_11-cuda-aarch6412_6-aarch64 (gh) (this job did not run on the merge base, so DrCI cannot tell whether the failure is pre-existing)
• Build Aarch64 Linux Wheels / pytorch/executorch / build-wheel-py3_11-cuda-aarch6413_0-aarch64 (gh) (this job did not run on the merge base, so DrCI cannot tell whether the failure is pre-existing)
• Build Aarch64 Linux Wheels / pytorch/executorch / build-wheel-py3_11-cuda-aarch6413_2-aarch64 (gh) (this job did not run on the merge base, so DrCI cannot tell whether the failure is pre-existing)
• Build Aarch64 Linux Wheels / pytorch/executorch / build-wheel-py3_12-cuda-aarch6412_6-aarch64 (gh) (this job did not run on the merge base, so DrCI cannot tell whether the failure is pre-existing)
• Build Aarch64 Linux Wheels / pytorch/executorch / build-wheel-py3_12-cuda-aarch6413_0-aarch64 (gh) (this job did not run on the merge base, so DrCI cannot tell whether the failure is pre-existing)
• Build Aarch64 Linux Wheels / pytorch/executorch / build-wheel-py3_12-cuda-aarch6413_2-aarch64 (gh) (this job did not run on the merge base, so DrCI cannot tell whether the failure is pre-existing)
• Build Aarch64 Linux Wheels / pytorch/executorch / build-wheel-py3_13-cuda-aarch6412_6-aarch64 (gh) (this job did not run on the merge base, so DrCI cannot tell whether the failure is pre-existing)
• Build Aarch64 Linux Wheels / pytorch/executorch / build-wheel-py3_13-cuda-aarch6413_0-aarch64 (gh) (this job did not run on the merge base, so DrCI cannot tell whether the failure is pre-existing)
• Build Aarch64 Linux Wheels / pytorch/executorch / build-wheel-py3_13-cuda-aarch6413_2-aarch64 (gh) (this job did not run on the merge base, so DrCI cannot tell whether the failure is pre-existing)
• Build Aarch64 Linux Wheels / pytorch/executorch / build-wheel-py3_14-cuda-aarch6412_6-aarch64 (gh) (this job did not run on the merge base, so DrCI cannot tell whether the failure is pre-existing)
• Build Aarch64 Linux Wheels / pytorch/executorch / build-wheel-py3_14-cuda-aarch6413_0-aarch64 (gh) (this job did not run on the merge base, so DrCI cannot tell whether the failure is pre-existing)
• Build Aarch64 Linux Wheels / pytorch/executorch / build-wheel-py3_14-cuda-aarch6413_2-aarch64 (gh) (this job did not run on the merge base, so DrCI cannot tell whether the failure is pre-existing)
• Build macOS Wheels / pytorch/executorch / build-wheel-py3_10-cpu (gh) (this job did not run on the merge base, so DrCI cannot tell whether the failure is pre-existing)
  Library not loaded: @rpath/libtorch_python.dylib
• Build macOS Wheels / pytorch/executorch / build-wheel-py3_11-cpu (gh) (this job did not run on the merge base, so DrCI cannot tell whether the failure is pre-existing)
  Library not loaded: @rpath/libtorch_python.dylib
• Build macOS Wheels / pytorch/executorch / build-wheel-py3_12-cpu (gh) (this job did not run on the merge base, so DrCI cannot tell whether the failure is pre-existing)
  Library not loaded: @rpath/libtorch_python.dylib
• Build macOS Wheels / pytorch/executorch / build-wheel-py3_13-cpu (gh) (this job did not run on the merge base, so DrCI cannot tell whether the failure is pre-existing)
  Library not loaded: @rpath/libtorch_python.dylib
• Build macOS Wheels / pytorch/executorch / build-wheel-py3_14-cpu (gh) (this job did not run on the merge base, so DrCI cannot tell whether the failure is pre-existing)
  pyarrow
• Build macOS Wheels / pytorch/executorch / upload / upload-wheel-py3_10-cpu (gh) (this job did not run on the merge base, so DrCI cannot tell whether the failure is pre-existing)
  Unable to download artifact(s): Artifact not found
• Build macOS Wheels / pytorch/executorch / upload / upload-wheel-py3_11-cpu (gh) (this job did not run on the merge base, so DrCI cannot tell whether the failure is pre-existing)
  Unable to download artifact(s): Artifact not found
• Build macOS Wheels / pytorch/executorch / upload / upload-wheel-py3_12-cpu (gh) (this job did not run on the merge base, so DrCI cannot tell whether the failure is pre-existing)
  Unable to download artifact(s): Artifact not found
• Build macOS Wheels / pytorch/executorch / upload / upload-wheel-py3_13-cpu (gh) (this job did not run on the merge base, so DrCI cannot tell whether the failure is pre-existing)
  Unable to download artifact(s): Artifact not found
• Build macOS Wheels / pytorch/executorch / upload / upload-wheel-py3_14-cpu (gh) (this job did not run on the merge base, so DrCI cannot tell whether the failure is pre-existing)
  Unable to download artifact(s): Artifact not found
• Build Windows Wheels / pytorch/executorch / build-wheel-py3_14-cpu (gh) (this job did not run on the merge base, so DrCI cannot tell whether the failure is pre-existing)
• nightly / test-cortex-m-e2e-nightly / run (deep_autoencoder, cortex-m55) / deep_autoencoder-cortex-m55 (gh) (this job did not run on the merge base, so DrCI cannot tell whether the failure is pre-existing)
  API rate limit exceeded
• Test Vulkan Backend (specialized runners) / build-vulkan-windows-msvc / windows-job (gh) (this job did not run on the merge base, so DrCI cannot tell whether the failure is pre-existing)
  ##[error]API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID C323:313069:1F9C727:6952A85:6A726A4E and timestamp 2026-08-04 22:40:14 UTC. For more on scraping GitHub and how it may affect your rights, please review our Terms of Service (https://docs.github.com/en/site-policy/github-terms/github-terms-of-service) - https://docs.github.com/en/rest/using-the-rest-api/getting-started-with-the-rest-api#rate-limiting
• trunk / test-models-windows (mobilebert, xnnpack-q8) / windows-job (gh) (this job did not run on the merge base, so DrCI cannot tell whether the failure is pre-existing)
  API rate limit exceeded
• trunk / test-models-windows (vit, xnnpack-q8) / windows-job (gh) (this job did not run on the merge base, so DrCI cannot tell whether the failure is pre-existing)
  ##[error]API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID C31F:6CF9E:20F2C41:6E24909:6A726A6F and timestamp 2026-08-04 22:40:47 UTC. For more on scraping GitHub and how it may affect your rights, please review our Terms of Service (https://docs.github.com/en/site-policy/github-terms/github-terms-of-service) - https://docs.github.com/en/rest/using-the-rest-api/getting-started-with-the-rest-api#rate-limiting
• Windows MSVC Build / build-windows-msvc / windows-job (gh) (this job did not run on the merge base, so DrCI cannot tell whether the failure is pre-existing)
  ##[error]API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID C32B:2C1547:9212D2:1EBB98E:6A726A24 and timestamp 2026-08-04 22:39:32 UTC. For more on scraping GitHub and how it may affect your rights, please review our Terms of Service (https://docs.github.com/en/site-policy/github-terms/github-terms-of-service) - https://docs.github.com/en/rest/using-the-rest-api/getting-started-with-the-rest-api#rate-limiting

FLAKY - The following jobs failed but were likely due to flakiness present on trunk:

• nightly / test-cortex-m-e2e-nightly / run (ds_cnn, cortex-m55) / ds_cnn-cortex-m55 (gh) (detected as infra flaky with no log or failing log classifier)
• nightly / test-cortex-m-e2e-nightly / run (mobilenet_v1_025, cortex-m55) / mobilenet_v1_025-cortex-m55 (gh) (detected as infra flaky with no log or failing log classifier)
• nightly / test-cortex-m-e2e-nightly / run (resnet8, cortex-m55) / resnet8-cortex-m55 (gh) (detected as infra flaky with no log or failing log classifier)
• pull / test-lora-linux / linux-job (gh) (detected as infra flaky with no log or failing log classifier)
• pull / test-lora-multimethod-linux / linux-job (gh) (detected as infra flaky with no log or failing log classifier)
• pull / test-models-linux (emformer_join, xnnpack-quantization-delegation, linux.4xlarge.memory) / linux-job (gh) (detected as infra flaky with no log or failing log classifier)
• pull / test-models-linux (ic4, xnnpack-quantization-delegation, linux.4xlarge.memory) / linux-job (gh) (detected as infra flaky with no log or failing log classifier)
• pull / test-phi-3-mini-runner-linux / linux-job (gh) (detected as infra flaky with no log or failing log classifier)
• trunk / test-arm-backend-ethos-u (test_ootb_tests_tosa) / linux-job (gh) (detected as infra flaky with no log or failing log classifier)
• trunk / test-arm-backend-ethos-u (test_pytest_models_ethos_u85) / linux-job (gh) (detected as infra flaky with no log or failing log classifier)
• trunk / test-arm-backend-ethos-u (test_run_ethos_u55) / linux-job (gh) (detected as infra flaky with no log or failing log classifier)
• trunk / test-arm-backend-ethos-u (test_smaller_stories_llama_tosa) / linux-job (gh) (detected as infra flaky with no log or failing log classifier)
• trunk / test-arm-backend-vkml (test_pytest_ops_vkml) / linux-job (gh) (detected as infra flaky with no log or failing log classifier)
• trunk / test-cortex-m-e2e / run (mv2, cortex-m55) / mv2-cortex-m55 (gh) (detected as infra flaky with no log or failing log classifier)
• trunk / test-cortex-m-e2e / run (mv3, cortex-m0plus) / mv3-cortex-m0plus (gh) (detected as infra flaky with no log or failing log classifier)
• trunk / test-huggingface-transformers-xnnpack (qwen3-1.7b|xnnpack|--quantize) / linux-job (gh) (detected as infra flaky with no log or failing log classifier)
• trunk / test-huggingface-transformers-xnnpack (smollm3-3b|xnnpack|--quantize) / linux-job (gh) (detected as infra flaky with no log or failing log classifier)
• trunk / test-models-windows (emformer_transcribe, portable) / windows-job (gh) (detected as infra flaky with no log or failing log classifier)
• trunk / test-models-windows (emformer_transcribe, xnnpack-q8) / windows-job (gh) (detected as infra flaky with no log or failing log classifier)
• trunk / test-models-windows (mobilebert, portable) / windows-job (gh) (detected as infra flaky with no log or failing log classifier)
• trunk / test-models-windows (mv3, portable) / windows-job (gh) (detected as infra flaky with no log or failing log classifier)
• trunk / test-models-windows (mv3, xnnpack-q8) / windows-job (gh) (detected as infra flaky with no log or failing log classifier)
• trunk / test-models-windows (resnet50, portable) / windows-job (gh) (detected as infra flaky with no log or failing log classifier)
• trunk / test-models-windows (resnet50, xnnpack-q8) / windows-job (gh) (detected as infra flaky with no log or failing log classifier)
• trunk / test-models-windows (vit, portable) / windows-job (gh) (detected as infra flaky with no log or failing log classifier)

BROKEN TRUNK - The following jobs failed but were present on the merge base:

👉 Rebase onto the `viable/strict` branch to avoid these failures

• pull / test-mcu-cortex-m-backend / linux-job (gh) (trunk failure)
• trunk / test-arm-backend-ethos-u (test_deit_e2e_ethos_u) / linux-job (gh) (trunk failure)
• trunk / test-arm-backend-ethos-u (test_memory_allocation) / linux-job (gh) (trunk failure)
• trunk / test-arm-backend-ethos-u (test_model_smollm2_135M_ethos_u85) / linux-job (gh) (trunk failure)
• trunk / test-arm-backend-ethos-u (test_pytest_models_ethos_u55) / linux-job (gh) (trunk failure)
• trunk / test-arm-backend-ethos-u (test_pytest_ops_ethos_u85) / linux-job (gh) (trunk failure)
• trunk / test-arm-backend-vkml (test_pytest_models_vkml) / linux-job (gh) (trunk failure)
• trunk / test-cortex-m-e2e / run (mv2, cortex-m0plus) / mv2-cortex-m0plus (gh) (trunk failure)
• trunk / test-cortex-m-e2e / run (mv2, cortex-m7) / mv2-cortex-m7 (gh) (trunk failure)
• trunk / test-cortex-m-e2e / run (mv3, cortex-m55) / mv3-cortex-m55 (gh) (trunk failure)
• trunk / test-cortex-m-ops / run (cortex-m0plus) / cortex-m-ops-cortex-m0plus (gh) (trunk failure)
• trunk / test-cortex-m-ops / run (cortex-m7) / cortex-m-ops-cortex-m7 (gh) (trunk failure)
• trunk / test-huggingface-transformers-xnnpack (gemma3-1b|xnnpack|--quantize) / linux-job (gh) (trunk failure)
• trunk / test-huggingface-transformers-xnnpack (llama3.2-1b|xnnpack|--quantize) / linux-job (gh) (trunk failure)
• trunk / test-huggingface-transformers-xnnpack (smollm2-135m|xnnpack|--quantize) / linux-job (gh) (trunk failure)

This comment was automatically generated by Dr. CI and updates every 15 minutes.