Weekly compatibility tests against oldest supported PyTorch - #745
Open
swahtz wants to merge 1 commit into
Open
Conversation
swahtz
marked this pull request as ready for review
August 20, 2026 01:56
swahtz
requested review from
areidmeyer and
matthewdcong
and removed request for
a team
August 20, 2026 01:56
Implements the weekly testing regime documented in docs/installation.rst (Notes on Testing, Compatibility, and Distribution): build and test main against the oldest PyTorch version that publishes wheels for the lowest CUDA version implied by the stable-PyTorch policy. For the PyTorch 2.13 matrix that is PyTorch 2.9.1 with CUDA 13.0. - .github/versions.json: record the tested versions under weekly_compat so they are updated alongside the rest of the version matrix. - load-versions.yml: expose compat-torch-version, compat-torch-full-version, compat-cuda-version, compat-cuda-tag, and compat-cuda-patch outputs. - weekly-compat.yml: new scheduled workflow (Saturdays 8am UTC, plus manual dispatch) mirroring the cu130 build/test pipeline without the PR-merge plumbing: EC2 CPU runner builds the wheel and gtests against the compat PyTorch/CUDA pin, then an EC2 GPU runner runs gtests and the Python unit tests. Doc tests are omitted since the documentation examples target the current in-development versions. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Signed-off-by: Jonathan Swartz <jonathan@jswartz.info>
swahtz
force-pushed
the
weekly-compat-tests
branch
from
August 20, 2026 03:37
86e763e to
a5858a1
Compare
swahtz
added
the
CI
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Implements the weekly testing regime described in the installation docs (Notes on Testing, Compatibility, and Distribution): test fvdb-core weekly against the oldest PyTorch version that publishes wheels for the lowest CUDA version implied by the "stable PyTorch's latest two CUDA versions" policy. For the PyTorch 2.13 matrix (#738) that is PyTorch 2.9.1 with CUDA 13.0 (verified against the versions published on download.pytorch.org/whl/cu130).
Changes:
.github/versions.json: newweekly_compatblock recording the tested torch/cuda versions, so they get reviewed and updated alongside the rest of the version matrix each cycle..github/workflows/load-versions.yml: new outputscompat-torch-version,compat-torch-full-version,compat-cuda-version,compat-cuda-tag, andcompat-cuda-patch(the container-image patch is looked up from the existingcuda.versionsmap, so the compat CUDA must be one of the supported versions — which the policy guarantees)..github/workflows/weekly-compat.yml: new scheduled workflow (Saturdays 8am UTC +workflow_dispatch) mirroring thecu130.ymlpipeline without thepull_request_targetmerge plumbing: provision EC2 CPU runner → build wheel/gtests against the compat PyTorch/CUDA pin → provision EC2 GPU runner → run gtests and Python unit tests → tear down runners.CHANGES.md: note the new weekly compatibility testing under 0.6.0.Notes:
pytest --markdown-docs) are intentionally omitted: the documentation examples pin the current in-development versions, not the compat versions.actionlint,zizmor --persona=regular, and the repo's EC2 runner-token policy check locally.weekly_compatvalues are a deliberate second source of truth fromtorch.version: at each new dev cycle (like PyTorch 2.13 #738), whoever bumps the matrix consults PyTorch's release compatibility matrix and updatesweekly_compatper the policy.🤖 Generated with Claude Code