Add kocommongen-v2

app.py

@@ -78,8 +78,8 @@ BENCHMARK_COLS = [
         AutoEvalColumn.hellaswag,
         AutoEvalColumn.mmlu,
         AutoEvalColumn.truthfulqa,
+        # AutoEvalColumn.commongen_v2,
         # TODO: Uncomment when we have results for these
-        # AutoEvalColumn.commongen,
         # AutoEvalColumn.ethicalverification,
     ]
 ]

src/display_models/read_results.py

@@ -9,14 +9,14 @@ import numpy as np
 from src.display_models.utils import AutoEvalColumn, make_clickable_model
 
 METRICS = ["acc_norm", "acc_norm", "acc", "mc2"]
-BENCHMARKS = ["ko_arc_challenge", "ko_hellaswag", "ko_mmlu", "ko_truthfulqa_mc"] #, "ko_commongen", "ethicalverification"]
+BENCHMARKS = ["ko_arc_challenge", "ko_hellaswag", "ko_mmlu", "ko_truthfulqa_mc", "ko_commongen_v2"] #, "ethicalverification"]
 BENCH_TO_NAME = {
     "ko_arc_challenge": AutoEvalColumn.arc.name,
     "ko_hellaswag": AutoEvalColumn.hellaswag.name,
     "ko_mmlu": AutoEvalColumn.mmlu.name,
     "ko_truthfulqa_mc": AutoEvalColumn.truthfulqa.name,
+    "ko_commongen_v2": AutoEvalColumn.commongen_v2.name,
     # TODO: Uncomment when we have results for these
-    # "ko_commongen": AutoEvalColumn.commongen.name,
     # "ethicalverification": AutoEvalColumn.ethicalverification.name,
 }
 

src/display_models/utils.py

@@ -29,8 +29,8 @@ class AutoEvalColumn:  # Auto evals column
     hellaswag = ColumnContent("Ko-HellaSwag", "number", True)
     mmlu = ColumnContent("Ko-MMLU", "number", True)
     truthfulqa = ColumnContent("Ko-TruthfulQA", "number", True)
+    commongen_v2 = ColumnContent("Ko-CommonGen V2", "number", True)
     # TODO: Uncomment when we have results for these
-    # commongen = ColumnContent("Ko-CommonGen", "number", True)
     # ethicalverification = ColumnContent("EthicalVerification", "number", True)
     model_type = ColumnContent("Type", "str", False)
     precision = ColumnContent("Precision", "str", False)  # , True)

src/load_from_hub.py

@@ -79,7 +79,7 @@ def get_leaderboard_df(
         all_data.append(gpt4_values)
         all_data.append(gpt35_values)
 
-    all_data.append(baseline)
+    # all_data.append(baseline)
     apply_metadata(all_data)  # Populate model type based on known hardcoded values in `metadata.py`
 
     df = pd.DataFrame.from_records(all_data)