Always use zero-shot results

parse_results.py

@@ -63,8 +63,9 @@ def parse_outputs(dataset_id: str) -> dict[str, Any]:
         df = ds[run].to_pandas()
         for task, result in json.loads(df.results.iloc[0]).items():
             if task != "all":
-                _, benchmark, _ = task.split("|")
-                metrics[benchmark] = result
+                _, benchmark, n_shots = task.split("|")
+                if int(n_shots) == 0:
+                    metrics[benchmark] = result
 
         versions.update(json.loads(df.versions.iloc[0]))