exp_rpt_softwareheritage-large-v2
收藏资源简介:
# exp_rpt_softwareheritage-large v2 Patched re-upload of `DCAgent/exp_rpt_softwareheritage-large` for the laion org. ## Patch summary v2: auto-install test deps in test.sh (drops only py_compile fails + non-PyPI imports); kept 2340/4998 tasks (46.8%) For each task in the source dataset, the patcher: 1. Runs `py_compile` on `tests/test_solution.py`; drops the task if it fails. 2. Parses top-level imports from the test file (AST: `import X`, `from X import …`). 3. Classifies each non-stdlib non-pre-installed import: - **PyPI-known** (the package exists on PyPI, possibly via a name map like `yaml→pyyaml`): generates a `pip install` line. - **Project-internal** (looks like a local helper module — e.g. `tests`, `src`, `helpers`, `model`): drops the task. - **Not on PyPI** (HEAD https://pypi.org/pypi/<n>/json returns 404): drops the task. 4. Injects the install lines into `tests/test.sh` immediately before the `pytest` invocation, wrapped in `|| true` so a single failed install does not kill the verifier. This supersedes the prior filter-only strategy that kept only 4.6% of tasks. ## Patcher The full patcher source is at `data/patchers/patch_softwareheritage_tasks.py` in the OpenThoughts-Agent repo.



