prompt: record follow-up outcomes for the retime-sweep bugs - #291
Merged
Conversation
- gradient_eager_jit_divergence_py313.md: root-cause UPDATE — not pure_callback (structurally absent from the rectangular path), not a jax version delta (both legs resolve jax 0.11.1); a one-iteration branch flip in the positive-only NNLS solver's data-dependent while_loop, whose ~1.577e-3 quantum matches the eager/jit gap to 5 significant figures. Guard is load-bearing; preferred fix is pinning nnls_solver_tol/max_iter. - delaunay_nan_probe_ell_comps_validator.md: status → fix shipped (probe reworked below the model-construction boundary). - complete/2026/08/smoke-surface-retime-sweep.md: the '900s jax_grad cap ~15x oversized' follow-up withdrawn — the cap's measured basis is the weekly channel where point_source/jax_grad/gradient.py runs at 568.2s. Co-Authored-By: Claude <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01EowgdevFczbMELBsoMWnsP
Jammy2211
pushed a commit
that referenced
this pull request
Aug 26, 2026
Jammy2211
pushed a commit
that referenced
this pull request
Aug 26, 2026
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01EqUT9Wc6mVLiJmbcstT7Z3
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
2 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Outcomes of the three follow-ups left open by
complete/2026/08/smoke-surface-retime-sweep.md:gradient_eager_jit_divergence_py313.mdgains a 193-line root-cause UPDATE: notpure_callback(structurally absent from the rectangular-mesh path), not a jax-version delta (both CI legs resolve identicaljax 0.11.1stacks) — a one-iteration branch flip in the positive-only NNLS solver's data-dependentlax.while_loop, whose ~1.577e-3 solver quantum matches the eager/jit gap to 5 significant figures across five independent FD back-outs. Theassert_eager_jit_consistentguard is load-bearing (AD runs on eager, FD on jitted — different NNLS branches would invalidate the certification), so no tolerance was widened; preferred fix is pinningnnls_solver_tol/nnls_max_itervia the already-plumbed Settings.delaunay_nan_probe_ell_comps_validator.mdstatus → fix shipped (probe reworked below the model-construction boundary inautolens_workspace_testbranchfix/delaunay-nan-probe; CI validation + PR in flight).jax_grad/cap" follow-up is withdrawn with the correcting evidence: the cap's measured basis is the weekly workspace-smoke channel, wherepoint_source/jax_grad/gradient.pyruns at 568.2 s (63 % of budget, run 30938311069) — live and unskipped. The 61 s figure that motivated the follow-up was a single different script.Generated by Claude Code