Skip to content

Question: intended path to run end-to-end evaluation (F2P / HP / EC / CFG)? #5

Description

@lapidshay

Thanks for releasing TestExplora, it's a really useful benchmark and I'm keen to build on it.

I'm trying to run the full pipeline end-to-end (generation -> execution on base/head -> the F2P/HP/EC/CFG metrics) and want to make sure I'm following the intended path rather than reinventing something you already provide - quite possibly I've just missed an entrypoint.

A couple of things I wasn't sure about:

  • I can see the generation side (inference.py) and the execution building blocks (test_runner.py, eval_docker.py, eval_ces.py, repo_manager.py), plus load_data4eva in data_manager.py. Is there a top-level driver that ties generation -> run on base_commit -> apply code_patch -> run on head -> compute/aggregate the metrics, or is that meant to be assembled by the user?

  • eval_ces.py / ces_api.py look like they target an internal execution service (CES). For running outside your environment, is eval_docker.py the intended path?

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions