feat: add functions to truncate numbers to one decimal place and format DataFrame columns accordingly

app.py

@@ -65,6 +65,29 @@ print("///// --- Settings --- /////", settings.model_dump())
 ) = get_evaluation_queue_df(settings.EVAL_REQUESTS_PATH, EVAL_COLS)
 
 
+def truncate_to_one_decimal(value):
+    """
+    将数字截断到1位小数（不四舍五入）
+    """
+    if pd.isna(value) or not isinstance(value, (int, float)):
+        return value
+    return float(int(value * 10)) / 10
+
+
+def format_dataframe_numbers(df: pd.DataFrame) -> pd.DataFrame:
+    """
+    格式化 DataFrame 中的数字列，只保留1位小数并截断
+    """
+    df = df.copy()
+    for col in df.columns:
+        if col in ['Model', 'T']:  # 跳过非数字列
+            continue
+        # 检查是否为数值类型
+        if pd.api.types.is_numeric_dtype(df[col]):
+            df[col] = df[col].apply(truncate_to_one_decimal)
+    return df
+
+
 def filter_dataframe_by_columns(selected_cols: list[str], original_df: pd.DataFrame) -> pd.DataFrame:
     """
     根据选择的列过滤 DataFrame
@@ -179,6 +202,8 @@ def init_leaderboard_tabs(
     precision_filtered_df = filter_dataframe_by_precision(default_precision, original_df)
     # 根据默认选择再筛选一次 DataFrame
     initial_filtered_df = filter_dataframe_by_columns(default_selected, precision_filtered_df)
+    # 格式化数字列，只保留1位小数并截断
+    initial_filtered_df = format_dataframe_numbers(initial_filtered_df)
 
     with gr.Row():
         with gr.Column(scale=1):
@@ -231,6 +256,8 @@ def init_leaderboard_tabs(
         column_filtered_df = filter_dataframe_by_columns(selected_cols, precision_filtered_df)
         # 最后按搜索关键词筛选
         final_df = search_models_in_dataframe(search_text, column_filtered_df)
+        # 格式化数字列，只保留1位小数并截断
+        final_df = format_dataframe_numbers(final_df)
         return final_df
 
     # 绑定搜索、列选择和 precision 的变化事件，动态更新 DataFrame

scripts/upload_dataset.py

@@ -1,6 +1,18 @@
-#!/usr/bin/env python3
+# !/usr/bin/env python3
+# /// script
+# dependencies = [
+#   "python-dotenv",
+#   "huggingface-hub",
+# ]
+# ///
 """
 Upload the eval-results/leaderboard folder to y-playground/results on Hugging Face Hub.
+
+Usage:
+
+```bash
+uv run scripts/upload_dataset.py
+```
 """
 
 import os
@@ -14,7 +26,9 @@ load_dotenv()
 
 # Configuration
 LOCAL_FOLDER = Path("eval-results/leaderboard")
-REPO_ID = "y-playground/results"
+HF_OWNER = os.getenv("HF_OWNER", "lmms-lab-si")
+HF_RESULTS_REPO_NAME = os.getenv("HF_RESULTS_REPO_NAME", "EASI-Leaderboard-Results")
+REPO_ID = f"{HF_OWNER}/{HF_RESULTS_REPO_NAME}"
 REPO_TYPE = "dataset"  # or "model" or "space"
 
 

src/populate.py

@@ -23,6 +23,15 @@ from src.display.utils import AutoEvalColumn, EvalQueueColumn
 from src.leaderboard.read_evals import get_raw_eval_results
 
 
+def truncate_to_one_decimal(value):
+    """
+    将数字截断到1位小数（不四舍五入）
+    """
+    if pd.isna(value) or not isinstance(value, (int, float)):
+        return value
+    return float(int(value * 10)) / 10
+
+
 def get_leaderboard_df(
     results_path: str,
     requests_path: str,
@@ -49,7 +58,7 @@ def get_leaderboard_df(
             exclude entries with missing benchmark results.
 
     Note:
-        The function automatically rounds numeric values to 2 decimal places and
+        The function automatically truncates numeric values to 1 decimal place and
         filters out any entries that have NaN values in the specified benchmark columns.
     """
     raw_data = get_raw_eval_results(results_path, requests_path)
@@ -57,7 +66,12 @@ def get_leaderboard_df(
 
     df = pd.DataFrame.from_records(all_data_json)
     df = df.sort_values(by=[AutoEvalColumn.average.name], ascending=False)
-    df = df.loc[:, cols].round(decimals=2)
+    df = df.loc[:, cols]
+
+    # 截断数字列到1位小数（不四舍五入）
+    for col in df.columns:
+        if col not in ['Model', 'T'] and pd.api.types.is_numeric_dtype(df[col]):
+            df[col] = df[col].apply(truncate_to_one_decimal)
 
     # filter out if any of the benchmarks have not been produced
     df = df.loc[has_no_nan_values(df, benchmark_cols), :]