Repository navigation
Reject unknown model_family; name the laser extra when prody is missing (#2484, #2483) - #212
Merged
Merged
Conversation
…is missing (#2484, #2483)
#2484: an unknown model_family silently fell through to stock ProteinMPNN. One
table in host/family_driver.py (STOCK_MODEL_FAMILIES + DRIVER_BACKED_FAMILY_MODULES)
now defines the valid names; validate_model_family enforces it in RunSpecification
__post_init__, the runner's _family_driver_for, prep, and the CLI --model-family
option (typer BadParameter listing the valid names). The default family is unchanged.
#2483: a lasermpnn request without the laser extra died on a bare
ModuleNotFoundError('prody') inside featurize. _require_family_extra now raises an
ImportError naming the extra (uv sync --extra laser / pip install 'aminx[laser]') at
family resolution, before featurization.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01SGHJhq24fYihaMivnXNhro
…h has no wheel -- main fails the same way) Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014bjCGQZgFEE6nR8cGTMFt7
…idate_model_family (CI regressions on #2484) The guard rejected families the codebase declares. ModelFamily (family_driver.py) is now the one Literal, adding "membrane" (named by the temperature-default rule and by MODEL_WEIGHTS). STOCK_MODEL_FAMILIES is derived from it minus the driver table, and RunSpecification annotates model_family with it, so the type and the runtime guard cannot drift. tests/host/test_comp536_campaign_manifest.py used "ligandmpnn_v1", which no table declares; it now uses the real family "ligandmpnn". Weight-variant names (soluble, sc, original, ligand) are deliberately not families and stay rejected. browser_validation_paths.json maps family_driver.py::validate_model_family to P26 (the RunSpecification surface), so the parity inventory has no unmapped symbol. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01SGHJhq24fYihaMivnXNhro
Owner
Author
|
Follow-up 36142c0 after CI caught a real regression: the first version's valid set {proteinmpnn, ligandmpnn, pottsmpnn, lasermpnn} rejected legitimate names. The valid set is now derived from one declared |
…amily tables, add protonpottsmpnn) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01SGHJhq24fYihaMivnXNhro
maraxen
added a commit
that referenced
this pull request
Oct 12, 2026
…y_wave: 9/9 would grade, 135/135 PASS) (#2621) Groups 1-4 ran sequentially on titanix at 43a7a3b (main 2acb6e7 with PRs #212-#215, plus the zscale single_mutation fold); every cool-tier record is completed/pass and every control failed. protonpotts_parity is the REVISION 261012 vehicle (99 controls). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01YGAcqn2SwMsYoLwzgobthN
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.
Fixes debts #2484 and #2483.
#2484: an unknown
model_familysilently ran stock ProteinMPNN. A typo such as--model-family foompnnfell through_family_driver_for(it returnedNonefor any name outside the driver table), so the run produced ProteinMPNN results under the wrong label. Valid names now live in one table insrc/aminx/host/family_driver.py(STOCK_MODEL_FAMILIES,DRIVER_BACKED_FAMILY_MODULES,MODEL_FAMILIES,validate_model_family). They are enforced at:RunSpecification.__post_init__(src/aminx/run/specs.py)_family_driver_for,src/aminx/host/runner.py)src/aminx/host/prep.py, before theFAMILY_DRIVERSlookup--model-familyoptions, viatyper.BadParameterAn unknown name raises
ValueError, naming it and listing the valid names. The default family is unchanged.#2483: a
lasermpnnrun without thelaserextra failed with a bareModuleNotFoundError('prody'). It now fails at family resolution, before featurisation, with anImportErrornaming the fix:uv sync --extra laserorpip install 'aminx[laser]'.Tests (
tests/host/test_model_family_validation.py): a single source of truth; valid names resolve; an unknown name is rejected through the spec, the runner and CLI parsing and invocation; a missingprodynames the extra and fails at resolution; withprodyimportable, no error. Run on titanix:tests/host -k "family or runner", 26 passed.🤖 Generated with Claude Code
https://claude.ai/code/session_01SGHJhq24fYihaMivnXNhro