release(0.1.19): pin torch<2.11 + GPU-compat self-check#30
Merged
Conversation
torch 2.11 dropped Volta (V100, CC 7.0) and Pascal (P100/GTX 10xx, CC 6.x) from its default cu128/cu129 wheels. With the previous <3 upper bound, `pip install scmidas` on those GPUs would silently install a broken torch — the first CUDA op would raise "no kernel image is available" from somewhere deep in the user's training loop. This patch: * tightens the upper bound to torch<2.11 (matching torchvision<0.26 and torchaudio<2.11) so default installs work on Volta/Pascal too; * adds a one-time GPU self-check at import that translates the cryptic CUDA error into actionable advice (downgrade torch or use cu126); * expands the CI matrix to test the new upper bound (torch 2.10 in place of the previous experimental "torch latest" job).
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
Pin
torch<2.11so defaultpip install scmidasworks on V100/Pascal GPUs.torch 2.11 dropped Volta (V100, CC 7.0) and Pascal (P100/GTX 10xx, CC 6.x) from its default cu128/cu129 wheels. Without this fix,
pip install scmidaswould silently install a torch that fails at the first CUDA op withno kernel image is available.Changes
pyproject.toml:torch<3→torch<2.11(+ matchingtorchvision<0.26/torchaudio<2.11)src/scmidas/__init__.py: one-time GPU self-check at import — translates the cryptic CUDA error into actionable guidance (downgrade torch or use cu126 wheel) for users who manually upgrade past the cap.github/workflows/test.yml: replace the experimentaltorch latestjob with a blockingtorch 2.10job (the new upper bound)pyproject.toml: bump to 0.1.19docs/source/release.md: 0.1.19 entryTest plan
pytest tests/— 37/37 passing locally on torch 2.8