Repository navigation
Build ggml GPU targets in compatible CI modes - #24223
Conversation
Summary: CI's default Linux configurations satisfy none of the CUDA, HIP or Metal ggml backend constraints, so CI only ever skipped the engine variants and runners that link them, and the CUDA and HIP op tests never ran in CI. These targets now take CI labels from `ggml_ci_labels`, loaded next to the ggml dependencies, which builds them where they are compatible. The open-source shim returns no labels, and the CPU variants keep the default configuration. The GPU engine variants and the Metal runner build for the host platform only, since their backends have no mobile flavors. Authored with Claude assistance. Differential Revision: D127883010
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/24223
Note: Links to docs will display an error until the docs builds have been completed. ❌ 1 New FailureAs of commit b141808 with merge base 8ecf0fb ( NEW FAILURE - The following job has failed:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
|
@digantdesai has exported this pull request. If you are a Meta employee, you can view the originating Diff in D127883010. |
This PR needs a
|
rascani
left a comment
There was a problem hiding this comment.
Review automatically exported from Phabricator review in Meta.
Summary:
CI's default Linux configurations satisfy none of the CUDA, HIP or Metal ggml backend constraints, so CI only ever skipped the engine variants and runners that link them, and the CUDA and HIP op tests never ran in CI.
These targets now take CI labels from
ggml_ci_labels, loaded next to the ggml dependencies, which builds them where they are compatible. The open-source shim returns no labels, and the CPU variants keep the default configuration. The GPU engine variants and the Metal runner build for the host platform only, since their backends have no mobile flavors.Authored with Claude assistance.
Differential Revision: D127883010