Skip to content

Build against latest cuOpt nightly wheels from RAPIDS nightly index - #23

Merged
0x17 merged 4 commits into
multigpu-pdlp-nightlyfrom
claude/cuopt-wheels-pypi-j6l929
Oct 7, 2026
Merged

0x17 merged 4 commits into
multigpu-pdlp-nightlyfrom
claude/cuopt-wheels-pypi-j6l929

Conversation

@0x17

@0x17 0x17 commented Oct 6, 2026

Copy link
Copy Markdown
Member

No description provided.

Re-applies 8edd916 (reverted in 4855ada) now that the split wheels
(libcuopt-mathopt-cuXX, libcuopt-client, libcuopt-routing-cuXX) are
published on the rapidsai nightly index, starting with 26.10.0a248.

Changes against the original attempt:
- Pin libcuopt-mathopt-cuXX and libcuopt-client directly instead of the
  libcuopt-cuXX umbrella. The umbrella only requires its parts as
  ==26.10.*, so pinning it no longer fixes the solver build that gets
  packaged.
- Drop the copies from libcuopt_client_cu12.libs / _cu13.libs. The
  client wheel carries no CUDA suffix and only ships
  libcuopt_client.libs, so those cp lines would fail the job.

Pinned to 26.10.0a250.post261006201351 (cuOpt main f123643).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01ALHyZUW9bDK51TntcQG4cg
@0x17
0x17 changed the base branch from main to multigpu-pdlp-nightly October 6, 2026 21:41
claude and others added 3 commits October 6, 2026 21:50
Nightly builds drop off the rapidsai index after a few days, so an
exact nightly pin breaks CI regularly. Constrain libcuopt-mathopt and
libcuopt-client to 26.10.* instead: CI follows the newest 26.10 nightly,
cannot jump to 26.12 nightlies from main, and picks up 26.10.0 final
once it is published. The exact cuOpt version and commit stay embedded
in the built link.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01ALHyZUW9bDK51TntcQG4cg
The CI runners have no GPU, so the gamslib regression test only runs on
developer machines. tests/colab-gpu-test.ipynb builds the link from a
chosen branch by running the run steps of main-x86_64.yml unchanged,
installs it into the downloaded GAMS system (picking CUDA 12 or 13 from
the driver), reads an optional GAMS license from the Colab secret
GAMS_LICENSE, and runs a trnsport smoke test that aborts on a
non-optimal solve, followed by tests/test-gamslib-cuopt.py.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01ALHyZUW9bDK51TntcQG4cg
@0x17
0x17 merged commit d917401 into multigpu-pdlp-nightly Oct 7, 2026
2 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants