Add default inference system prompt

pointerbench-sheets/eval.py

@@ -14,6 +14,7 @@ Accepted point keys: "point" / "pred" / "coordinate", or flat "x" and "y".
 Coordinates are absolute pixels on the 1024x768 image.
 
 Usage:
+    python eval.py --show-system-prompt
     python eval.py --predictions preds.jsonl
     python eval.py --predictions preds.jsonl --json report.json
 """
@@ -28,6 +29,15 @@ from pathlib import Path
 ROOT = Path(__file__).resolve().parent
 GT_PATH = ROOT / "data" / "test" / "metadata.jsonl"
 
+DEFAULT_SYSTEM_PROMPT = (
+    "You are evaluating Pointerbench, a GUI grounding benchmark. "
+    "You will receive one 1024x768 screenshot and one task instruction. "
+    "Use absolute pixel coordinates with origin at the top-left of the image. "
+    "Do not return normalized coordinates. Do not crop or resize the coordinate frame. "
+    "For point tasks, return JSON like {\"point\": [x, y]}. "
+    "For bounding-box tasks, return JSON like {\"bbox\": [x0, y0, x1, y1]}."
+)
+
 
 def _load_jsonl(path: Path) -> list[dict]:
     text = path.read_text(encoding="utf-8").strip()
@@ -108,7 +118,9 @@ def _print(report: dict) -> None:
 def main() -> None:
     ap = argparse.ArgumentParser(description=__doc__,
                                  formatter_class=argparse.RawDescriptionHelpFormatter)
-    ap.add_argument("--predictions", required=True, type=Path,
+    ap.add_argument("--show-system-prompt", action="store_true",
+                    help="print the recommended inference system prompt and exit")
+    ap.add_argument("--predictions", type=Path,
                     help="JSONL/JSON predictions: {id, point:[x,y]} per example")
     ap.add_argument("--gt", type=Path, default=GT_PATH,
                     help=f"ground-truth metadata (default: {GT_PATH})")
@@ -116,6 +128,12 @@ def main() -> None:
                     help="also write the full report to this JSON path")
     args = ap.parse_args()
 
+    if args.show_system_prompt:
+        print(DEFAULT_SYSTEM_PROMPT)
+        return
+    if args.predictions is None:
+        ap.error("--predictions is required unless --show-system-prompt is used")
+
     gt = _load_jsonl(args.gt)
     preds = {r["id"]: r for r in _load_jsonl(args.predictions) if "id" in r}
     report = evaluate(gt, preds)