Repository navigation
On a branch cut from origin/main, commits c078d05 (origin/goal/13-no-python-file-under-artifactb: external assets by fla - #14
Merged
Conversation
The installable package hard-coded locations that exist only on the maintainer's machine, so `--model fst` (RESULTS_8WAY.md) could not run from a fresh clone and the v1.2 tools silently looked in someone else's home directory. - models/fst.py: FSTModel takes the clone from fst_repo or ARTIFACTBENCH_FST_REPO and raises a ValueError naming both when neither is set (no default). - bench.py: --fst-repo, --fst-stage1-ckpt, --fst-stage2-ckpt wired into FSTModel; the parser moved into build_parser() so the wiring is unit-tested. - v12/model_assets.py: asset_location() no longer has maintainer defaults for the ArtifactNet, CLAM and FST trees; unset variables raise with their name. - v12/freeze.py: --artifactnet-repo / ARTIFACTBENCH_ARTIFACTNET_REPO for the exposure audit (content hashes still pin the inputs). - rerun_current.py: --artifactnet-repo and --fma-full-root (env ARTIFACTBENCH_ARTIFACTNET_REPO / ARTIFACTBENCH_FMA_FULL_ROOT); the fma root is only required with --recover-fma-from. - RESULTS_8WAY.md: fenced 8-way command with --fst-repo; check_docs.py now covers RESULTS_8WAY.md (links and flags); README and REPRODUCING_v1.2.md updated. - tests/test_external_asset_locations.py: FST error path, env/arg precedence, CLI wiring, pinned-asset and rerun errors, and a scan that fails on any /home|/Users|/media|/Volumes|/mnt/<name>/ literal under artifactbench/ and scripts/. Assumption: maintainer workflows keep working by exporting the variables already documented in docs/REPRODUCING_v1.2.md; no behaviour beyond the location lookup changed. Historical command transcripts in docs/ARTIFACTNET_FROZEN_REPLAY_v1.2.md and paper/README.md still show maintainer paths; they are records of past runs, outside this goal's scope (Python files only).
It links docs/MUSEBENCH.md, states the $MUSEBENCH_ROOT data-root rule (nothing derived from MuseBench is committed; only the scripts are), and lists the stage commands in fenced blocks: index_shards, select (plan/manifest/eval-manifest), fetch, leak_map, verify (index/acquire/leak/all), report and python -m artifactbench.bench. scripts/check_docs.py accepts every flag (0 problems). Assumption: the runner example uses --model artifactnet, the public maintained detector already used throughout this README, because --model only accepts registered choices; the section names no scores and no private checkpoint.
unohee
added a commit
that referenced
this pull request
Nov 5, 2026
…in/go") pass again: could not make a fresh checkout (#15) * Locate external assets by flag or env var, never by a maintainer path The installable package hard-coded locations that exist only on the maintainer's machine, so `--model fst` (RESULTS_8WAY.md) could not run from a fresh clone and the v1.2 tools silently looked in someone else's home directory. - models/fst.py: FSTModel takes the clone from fst_repo or ARTIFACTBENCH_FST_REPO and raises a ValueError naming both when neither is set (no default). - bench.py: --fst-repo, --fst-stage1-ckpt, --fst-stage2-ckpt wired into FSTModel; the parser moved into build_parser() so the wiring is unit-tested. - v12/model_assets.py: asset_location() no longer has maintainer defaults for the ArtifactNet, CLAM and FST trees; unset variables raise with their name. - v12/freeze.py: --artifactnet-repo / ARTIFACTBENCH_ARTIFACTNET_REPO for the exposure audit (content hashes still pin the inputs). - rerun_current.py: --artifactnet-repo and --fma-full-root (env ARTIFACTBENCH_ARTIFACTNET_REPO / ARTIFACTBENCH_FMA_FULL_ROOT); the fma root is only required with --recover-fma-from. - RESULTS_8WAY.md: fenced 8-way command with --fst-repo; check_docs.py now covers RESULTS_8WAY.md (links and flags); README and REPRODUCING_v1.2.md updated. - tests/test_external_asset_locations.py: FST error path, env/arg precedence, CLI wiring, pinned-asset and rerun errors, and a scan that fails on any /home|/Users|/media|/Volumes|/mnt/<name>/ literal under artifactbench/ and scripts/. Assumption: maintainer workflows keep working by exporting the variables already documented in docs/REPRODUCING_v1.2.md; no behaviour beyond the location lookup changed. Historical command transcripts in docs/ARTIFACTNET_FROZEN_REPLAY_v1.2.md and paper/README.md still show maintainer paths; they are records of past runs, outside this goal's scope (Python files only). * README.md has a "Zero-shot evaluation on MuseBench" section It links docs/MUSEBENCH.md, states the $MUSEBENCH_ROOT data-root rule (nothing derived from MuseBench is committed; only the scripts are), and lists the stage commands in fenced blocks: index_shards, select (plan/manifest/eval-manifest), fetch, leak_map, verify (index/acquire/leak/all), report and python -m artifactbench.bench. scripts/check_docs.py accepts every flag (0 problems). Assumption: the runner example uses --model artifactnet, the public maintained detector already used throughout this README, because --model only accepts registered choices; the section names no scores and no private checkpoint.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Milestone of an auto-claude run.
Check:
export XDG_RUNTIME_DIR="${XDG_RUNTIME_DIR:-/run/user/$(id -u)}" && systemd-run --user --scope -q -p MemoryMax=6G ~/RTX_ENV/bin/python -m pytest -m "not slow" -q -p no:cacheprovider && ~/RTX_ENV/bin/python scripts/check_docs.py && ! grep -rnE "/home/unohee|/media/unohee" artifactbench scripts --include='*.py' && ~/RTX_ENV/bin/python -m artifactbench.bench --help | grep -q -- "--fst-repo" && grep -q "docs/MUSEBENCH.md" README.md && grep -q "MUSEBENCH_ROOT" README.md && test -f tests/test_external_asset_locations.py && ! grep -niE "jupiter|h5_soup|codec4" README.mdPlan: .claude/auto-plan.md (local to the run).