Skip to content

Loosen lbfgsb_test Rosenbrock tolerance to 1.5e-3#641

Open
seantalts wants to merge 1 commit into
google:mainfrom
seantalts:fix-lbfgsb-test-tolerance
Open

Loosen lbfgsb_test Rosenbrock tolerance to 1.5e-3#641
seantalts wants to merge 1 commit into
google:mainfrom
seantalts:fix-lbfgsb-test-tolerance

Conversation

@seantalts
Copy link
Copy Markdown
Contributor

@seantalts seantalts commented Jun 4, 2026

Update tolerance - a recent XLA:CPU change to strength reduction of batched vector vector products made this fail due to fp noise.

The L-BFGS-B test asserts the Rosenbrock function value is <= 1e-3 at
the optimizer's output, but this is flaky across platforms/backends
where the optimizer terminates just shy of that bound. Loosen the
tolerance to 1.5e-3.
@seantalts
Copy link
Copy Markdown
Contributor Author

@vroulet could you take a look? Small test-tolerance bump — a recent XLA:CPU strength-reduction change to batched vector-vector products introduced fp noise that pushes the Rosenbrock value just over the old 1e-3 bound.

@vroulet
Copy link
Copy Markdown
Collaborator

vroulet commented Jun 4, 2026

The library is no longer maintained, so failing test do not matter I believe. Why do you ask?

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