| |
| """Run both RE-1 executions and write the capsule metrics file. |
| |
| This is the single published top-level command for the experiment: |
| python3 scripts/run_capsule.py --output-dir /path/to/out |
| |
| It shells out to scripts/run_one.sh twice (fresh containers, same commands), |
| then runs harness/analyze.py and reduces the analysis to metrics.json. |
| """ |
|
|
| from __future__ import annotations |
|
|
| import argparse |
| import json |
| import os |
| import shutil |
| import subprocess |
| import sys |
| from pathlib import Path |
|
|
| HERE = Path(__file__).resolve().parent.parent |
|
|
|
|
| def main() -> int: |
| parser = argparse.ArgumentParser(description=__doc__) |
| parser.add_argument("--output-dir", required=True, type=Path) |
| parser.add_argument( |
| "--evidence-dir", |
| type=Path, |
| default=Path(os.environ.get("RE1_EVIDENCE_DIR", str(Path.cwd() / "evidence"))), |
| ) |
| args = parser.parse_args() |
|
|
| out = args.output_dir.resolve() |
| out.mkdir(parents=True, exist_ok=True) |
| env = {**os.environ, "RE1_EVIDENCE_DIR": str(args.evidence_dir)} |
|
|
| for run_id in ("r1", "r2"): |
| print(f"== {run_id} ==", flush=True) |
| subprocess.run(["bash", str(HERE / "scripts" / "run_one.sh"), run_id], check=True, env=env) |
|
|
| runs = args.evidence_dir / "runs" |
| proc = subprocess.run( |
| [ |
| sys.executable, |
| str(HERE / "harness" / "analyze.py"), |
| "--run1", str(runs / "r1"), |
| "--run2", str(runs / "r2"), |
| "--out", str(out / "analysis.json"), |
| "--md", str(out / "analysis.md"), |
| ] |
| ) |
| analysis = json.loads((out / "analysis.json").read_text()) |
|
|
| for run_id in ("r1", "r2"): |
| dest = out / run_id |
| dest.mkdir(parents=True, exist_ok=True) |
| shutil.copy2(runs / run_id / "results.json", dest / "results.json") |
| shutil.copy2(runs / run_id / "run_status.json", dest / "run_status.json") |
|
|
| def repro(key: str) -> float: |
| return analysis["quantities"].get(key, {}).get("rel_diff", 1.0) |
|
|
| metrics = { |
| "decision": analysis["decision"], |
| "max_relative_deviation": max( |
| (row["rel_diff"] for row in analysis["quantities"].values()), default=1.0 |
| ) |
| if not analysis["missing"] |
| else 1.0, |
| "missing_quantities": len(analysis["missing"]), |
| "failing_quantities": len(analysis["failing"]), |
| "quantities_checked": len(analysis["quantities"]), |
| "qwen_ppl_delta_repro": repro("qwen2.5-0.5b-instruct.ppl_delta"), |
| "qwen_hellaswag_delta_repro": repro("qwen2.5-0.5b-instruct.hellaswag_delta_pct"), |
| "smollm_ppl_delta_repro": repro("smollm2-360m-instruct.ppl_delta"), |
| "smollm_hellaswag_delta_repro": repro("smollm2-360m-instruct.hellaswag_delta_pct"), |
| } |
| (out / "metrics.json").write_text(json.dumps(metrics, indent=2, sort_keys=True) + "\n") |
| print(json.dumps(metrics, indent=2)) |
| return proc.returncode |
|
|
|
|
| if __name__ == "__main__": |
| raise SystemExit(main()) |
|
|
