SWE-rebench V2: Python Execution Traces for Code World Modeling

SWE-rebench V2 — CodeWorldModeling Traces is a derived dataset containing line-by-line Python execution traces for test suites. The dataset was produced from instances of nebius/SWE-rebench-V2 and is authored by marin-community. It was last updated on 2026-05-21.

Use Cases

• Training models for code execution prediction based on line-by-line trace data.
• Analyzing test suite behavior and coverage using captured execution traces.
• Developing debugging or fault localization tools based on runtime execution patterns.

Strengths

• Traces are captured from running tests inside controlled Nebius ConTree sandboxes, suggesting a reproducible environment.
• Each instance is described as coming with a fix, implying a paired problem-solution structure.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

Derived from instances of nebius/SWE-rebench-V2.

Collection Method

Traces captured by running each instance's tests under a tracer inside Nebius ConTree sandboxes.

Freshness

Last updated 2026-05-21 02:44:45; freshness should be verified.