diff --git a/evaluation/swe_bench/run_infer.py b/evaluation/swe_bench/run_infer.py index 8bd23939dd58..a171d5f20a56 100644 --- a/evaluation/swe_bench/run_infer.py +++ b/evaluation/swe_bench/run_infer.py @@ -91,6 +91,14 @@ def get_instruction(instance: pd.Series, metadata: EvalMetadata): "Your thinking should be thorough and so it's fine if it's very long.\n" ) + instruction += ( + '\n' + '- You MUST generate only one action per turn!\n' + '- A patch is a set of changes to the source code of the codebase that you are given\n' + '- You MUST generate a patch that attempts to fix the issue described in the \n' + '\n' + ) + if RUN_WITH_BROWSING: instruction += ( '\n'