Skip to content
Closed
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
41 changes: 41 additions & 0 deletions entries/swegym-lite.yaml
Original file line number Diff line number Diff line change
@@ -0,0 +1,41 @@
id: swegym-lite
name: "SWE-Gym Lite"
version: "0.1.0"
description: >
SWE-Gym Lite ported to the CUBE protocol — the 230-instance subset of
SWE-Gym (ICML 2025), a training environment of real-world Python
software-engineering tasks collected with the SWE-bench methodology across
11 repositories (moto, mypy, conan, dvc, dask, pydantic, pandas, hydra,
bokeh, MONAI, modin). The agent receives a GitHub issue and a git checkout
at the base commit inside a pre-built executable Docker container, and must
produce a patch that makes the upstream fail_to_pass tests pass without
breaking pass_to_pass.
package: swegym-lite-cube
# NOTE: points at the cube-harness feature branch until the cube-harness PR
# merges to the default branch; drop the @ref (or pin @dev) after merge.
dev_install_url: "git+https://github.com/The-AI-Alliance/cube-harness@feat/swegym-lite-cube#subdirectory=cubes/swegym-lite-cube"

authors:
- github: recursix
name: Alexandre Lacoste

legal:
wrapper_license: Apache-2.0
benchmark_license:
reported: Apache-2.0
source_url: "https://github.com/SWE-Gym/SWE-Gym/blob/main/LICENSE"
verified_by_original_authors: false
notices:
- type: third_party_data
description: >
Wraps the third-party SWE-Gym Lite dataset
(https://huggingface.co/datasets/SWE-Gym/SWE-Gym-Lite) and the
community-published per-instance evaluation images under the
xingyaoww/sweb.eval.x86_64.* Docker Hub namespace.

paper: "https://arxiv.org/abs/2412.21139"
getting_started_url: "https://github.com/SWE-Gym/SWE-Gym"
tags:
- coding
- swe
- docker
Loading