real leaderboard

app.py

@@ -14,7 +14,7 @@ from designs_submission_validations import validate_github_link, validate_user_d
 from domain_constants import SCORE_NAMES_MAP, USER_GEN_DESIGNS_COLUMNS
 
 sys.path.append(os.path.join(os.path.dirname(__file__), "bike_bench_internal/src/"))
-from bikebench.benchmarking import benchmarking_utils
+from bikebench.benchmarking.benchmarking_utils import evaluate_designs
 
 
 def compute_scores(user_gen_designs: pd.DataFrame) -> ModelScoringResult:
@@ -23,8 +23,7 @@ def compute_scores(user_gen_designs: pd.DataFrame) -> ModelScoringResult:
     if designs_length < 10_000:
         raise Exception(f"Too few designs to evaluate. Expected > 10,000, got {designs_length}")
     data_tens = torch.tensor(user_gen_designs.values, dtype=torch.float32)
-    main_scores, detailed_scores, all_evaluation_scores = benchmarking_utils.evaluate(data_tens, device="cpu",
-                                                                                      evaluate_as_aggregate=False)
+    main_scores, detailed_scores, all_evaluation_scores = evaluate_designs(data_tens)
     return ModelScoringResult(
         uuid=str(uuid.uuid4()),
         submission_time=datetime.datetime.now(),