SWE-agent takes a GitHub issue and tries to automatically fix it, using your LM of choice. It can also be employed for offensive cybersecurity or competitive coding challenges. [NeurIPS 2024]
Enh: Live output from sb-cli
Kilian Lieretf536fe0474c6
diff --git a/sweagent/run/hooks/swe_bench_evaluate.py b/sweagent/run/hooks/swe_bench_evaluate.pyindex 589e0cfc..55b2ebf4 100644--- a/sweagent/run/hooks/swe_bench_evaluate.py+++ b/sweagent/run/hooks/swe_bench_evaluate.py@@ -1,4 +1,5 @@import subprocess+import sysfrom pathlib import Pathfrom sweagent.run.hooks.abstract import RunHook@@ -16,8 +17,9 @@ class SweBenchEvaluate(RunHook):self.logger = get_logger("SB-evaluate", emoji="😬")def on_end(self) -> None:+ self.logger.info("Submitting results to SWE-Bench")try:- subprocess.check_output(+ subprocess.run(["sb-cli","submit",@@ -27,7 +29,10 @@ class SweBenchEvaluate(RunHook):self.output_dir / "preds.json","--run_id",self.output_dir.name,- ]+ ],+ check=True,+ stdout=sys.stdout,+ stderr=sys.stderr,)except subprocess.CalledProcessError as e:self.logger.error("Failed to submit results to SweBench eval: %s", e)