Files

14 lines
703 B
JSON

{
"schema_version": 1,
"status": "completed",
"experiment_id": "baseline-swebench",
"purpose": "No-communication baseline on a fixed 20-task ImpossibleBench SWE holdout for three models.",
"remote_docker_host": "ssh://[email protected]",
"model_runs": [
{"name": "glm", "plan": "experiments/baseline-swebench/glm-plan.json", "run_dir": "logs/baseline-swebench/glm/run"},
{"name": "muse", "plan": "experiments/baseline-swebench/muse-plan.json", "run_dir": "logs/baseline-swebench/muse/run"},
{"name": "deepseek", "plan": "experiments/baseline-swebench/deepseek-plan.json", "run_dir": "logs/baseline-swebench/deepseek/run-v2"}
],
"report_dir": "logs/baseline-swebench/report"
}