Fix PyTorch dependency and Average column KeyError

requirements.txt

@@ -16,5 +16,6 @@ tokenizers>=0.15.0
 sentencepiece
 
 # Additional dependencies for French medical NER
+torch>=2.6.0
 seqeval>=1.2.2
 scikit-learn>=1.3.0

src/display/utils.py

@@ -26,7 +26,6 @@ auto_eval_column_dict = []
 auto_eval_column_dict.append(["model_type_symbol", ColumnContent, ColumnContent("T", "str", True, never_hidden=True)])
 auto_eval_column_dict.append(["model", ColumnContent, ColumnContent("Model", "markdown", True, never_hidden=True)])
 #Scores
-auto_eval_column_dict.append(["average", ColumnContent, ColumnContent("Average ⬆️", "number", True)])
 for task in Tasks:
     auto_eval_column_dict.append([task.name, ColumnContent, ColumnContent(task.value.col_name, "number", True)])
 # Model information

src/populate.py

@@ -6,6 +6,7 @@ import pandas as pd
 from src.display.formatting import has_no_nan_values, make_clickable_model
 from src.display.utils import AutoEvalColumn, EvalQueueColumn
 from src.leaderboard.read_evals import get_raw_eval_results
+from src.about import Tasks
 
 
 def get_leaderboard_df(results_path: str, requests_path: str, cols: list, benchmark_cols: list) -> pd.DataFrame:
@@ -14,7 +15,9 @@ def get_leaderboard_df(results_path: str, requests_path: str, cols: list, benchm
     all_data_json = [v.to_dict() for v in raw_data]
 
     df = pd.DataFrame.from_records(all_data_json)
-    df = df.sort_values(by=[AutoEvalColumn.average.name], ascending=False)
+    # Sort by the first task (EMEA NER) since we don't have an average for NER tasks
+    first_task = list(Tasks)[0]  # emea_ner
+    df = df.sort_values(by=[getattr(AutoEvalColumn, first_task.name).name], ascending=False)
     df = df[cols].round(decimals=2)
 
     # filter out if any of the benchmarks have not been produced