prompt: retire interferometer OOM prompt, issue JAX traceback harness fix (#186) - #347
Merged
Merged
Conversation
… fix (#186) The OOM in draft/bug/autolens/interferometer_release_leg_oom.md was already root-caused and shipped before that prompt's own 2026-08-04 amendment was written: MultiStartProdigy ran 48 starts as one unbatched vmap (~86 GB in a single allocation), autolens_workspace#450 set batch_size=4, and Release Integrate run 30901054267 discharged the leg on 2026-08-04 (complete/2026/08/interferometer-start-here-integrate-oom.md). Re-verified before retiring: autolens_workspace and autogalaxy_workspace both carry batch_size=4 on the 48-start MultiStartProdigy in scripts/interferometer/start_here.py, and the interferometer shard is green in the two most recent nightly Release Integrate runs (32804373015 on 08-25, 33073386315 on 08-27). The prompt's NUFFT/transformer hypothesis was wrong. Retired to complete/archive/shelved/ with a superseded-by note rather than issued. The one item of the amendment that was never done — JAX_TRACEBACK_FILTERING=off in the release harness, still absent from PyAutoHeart at d576003 — is split out as its own task and issued as PyAutoLabs/PyAutoHeart#186. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01SFoQCsxRTGjKk2YGe4R9Rs
…ness Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01SFoQCsxRTGjKk2YGe4R9Rs
…#187) PyAutoHeart#187 merged as c8b093e1; issue #186 closed. Record written, prompt folded out of active/, complete/index.md refreshed, dashboard regenerated. Also swept draft/triage/nightly_release_blocked_eight_nights.md, whose 07-30 and 07-31 rows and "What is actually left" list still pointed at the now-retired draft/bug/autolens/interferometer_release_leg_oom.md as live work. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01SFoQCsxRTGjKk2YGe4R9Rs
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.
Mind-state half of a
/start_devrun ondraft/bug/autolens/interferometer_release_leg_oom.md. The prompt turned out to be stale, so nothing was issued for it as written.The prompt was already shipped
complete/2026/08/interferometer-start-here-integrate-oom.mdrecords the root cause and the fix, both from 2026-07-31 — before the prompt's own 2026-08-04 amendment was written:MultiStartProdigyran 48 starts as a single unbatchedvmap. 48 divides85898814480exactly (≈1.79 GB per start). Not the NUFFT/transformer, which the prompt's "Notes for triage" had raised the prior on.autolens_workspace#450setbatch_size=4; Release Integrate run 30901054267 discharged the leg on 2026-08-04 withscripts/interferometer/start_here.py ... PASS (173.5s).Re-verified before retiring, not taken from the record:
autolens_workspaceandautogalaxy_workspaceboth carrybatch_size=4on the 48-startMultiStartProdigyatscripts/interferometer/start_here.py:307— so the amendment's "it is not autolens-specific" is closed too.autogalaxy) and 33073386315 (2026-08-27,autolens). The 08-25multi_datasetfailure in that run ishttp.client.RemoteDisconnected, not an OOM.Retired to
complete/archive/shelved/with a superseded-by note rather than issued.What was issued instead
One item of the amendment was never done, and is still undone:
JAX_TRACEBACK_FILTERINGappears nowhere in PyAutoHeart. Split out as its own prompt and issued as PyAutoLabs/PyAutoHeart#186.Changes
draft/bug/autolens/interferometer_release_leg_oom.md→complete/archive/shelved/active/jax_traceback_filtering_release_harness.mdactive.mdjax-traceback-filtering-release-harness→ PyAutoHeart#186dashboard.md,dashboard.htmlpyauto-brain intake --apply dashboard)lifecycle.py check→ OK. The threedateswarnings ondraft/maintenance/ci/,draft/test/pyautoheart/anddraft/test/workspaces/prompts are pre-existing and untouched.Not done
12 workspace scripts still call
MultiStartProdigywith nobatch_size(autolens:group/,point_source/,multi_dataset/,cluster/,weak/, …; autogalaxy:multi_galaxy/modeling.py,multi_dataset/start_here.py,cluster/modeling.py). Cheaper datasets than a 108k-visibility fit, so likely fine — flagged, deliberately not filed.Generated by Claude Code