consolidate

app.py

@@ -43,47 +43,138 @@ def init_leaderboard(dataframe):
 def refresh_leaderboard():
     import sys
     import traceback
+    import pandas as pd
     
     try:
+        sys.stderr.write("=== REFRESH LEADERBOARD DEBUG ===\n")
         sys.stderr.write("Refreshing leaderboard data...\n")
         sys.stderr.flush()
         
         # Get fresh leaderboard data
         df = get_leaderboard_df(EVAL_RESULTS_PATH, COLS, BENCHMARK_COLS)
-        sys.stderr.write(f"Got DataFrame with shape: {df.shape}\n")
-        sys.stderr.write(f"DataFrame columns: {df.columns.tolist()}\n")
+        
+        sys.stderr.write(f"get_leaderboard_df returned: {type(df)}\n")
+        if df is not None:
+            sys.stderr.write(f"DataFrame shape: {df.shape}\n")
+            sys.stderr.write(f"DataFrame columns: {df.columns.tolist()}\n")
+            sys.stderr.write(f"DataFrame empty: {df.empty}\n")
+        else:
+            sys.stderr.write("DataFrame is None!\n")
         sys.stderr.flush()
         
         # Check if DataFrame is valid for leaderboard
         if df is None:
-            sys.stderr.write("DataFrame is None, cannot create leaderboard\n")
+            sys.stderr.write("DataFrame is None, creating fallback DataFrame\n")
             sys.stderr.flush()
-            raise ValueError("DataFrame is None")
+            # Create a fallback DataFrame
+            df = create_fallback_dataframe()
             
-        if df.empty:
-            sys.stderr.write("DataFrame is empty, creating minimal valid DataFrame\n")
+        elif df.empty:
+            sys.stderr.write("DataFrame is empty, creating fallback DataFrame\n")
+            sys.stderr.flush()
+            # Create a fallback DataFrame for empty case
+            df = create_fallback_dataframe()
+        
+        elif not all(col in df.columns for col in COLS):
+            sys.stderr.write(f"DataFrame missing required columns. Has: {df.columns.tolist()}, Needs: {COLS}\n")
             sys.stderr.flush()
-            # Create a minimal valid DataFrame that won't crash the leaderboard
-            import pandas as pd
-            empty_df = pd.DataFrame(columns=COLS)
-            # Add one dummy row to prevent leaderboard component from crashing
-            dummy_row = {col: 0 if col in BENCHMARK_COLS or col == AutoEvalColumn.average.name else "" for col in COLS}
-            dummy_row[AutoEvalColumn.model.name] = "No models evaluated yet"
-            dummy_row[AutoEvalColumn.model_type_symbol.name] = "?"
-            empty_df = pd.DataFrame([dummy_row])
-            return init_leaderboard(empty_df)
+            # Create a fallback DataFrame for missing columns
+            df = create_fallback_dataframe()
+        
+        sys.stderr.write(f"Final DataFrame for leaderboard - Shape: {df.shape}, Columns: {df.columns.tolist()}\n")
+        sys.stderr.flush()
         
-        sys.stderr.write("Creating leaderboard with valid DataFrame\n")
+        # Ensure DataFrame has the exact columns expected
+        for col in COLS:
+            if col not in df.columns:
+                sys.stderr.write(f"Adding missing column: {col}\n")
+                if col in BENCHMARK_COLS or col == AutoEvalColumn.average.name:
+                    df[col] = 0.0
+                elif col == AutoEvalColumn.model.name:
+                    df[col] = "Unknown Model"
+                elif col == AutoEvalColumn.model_type_symbol.name:
+                    df[col] = "?"
+                else:
+                    df[col] = ""
+                sys.stderr.flush()
+        
+        # Reorder columns to match expected order
+        df = df[COLS]
+        
+        sys.stderr.write("Creating leaderboard component...\n")
         sys.stderr.flush()
-        return init_leaderboard(df)
+        
+        new_leaderboard = init_leaderboard(df)
+        sys.stderr.write("Leaderboard component created successfully\n")
+        sys.stderr.flush()
+        
+        return new_leaderboard
         
     except Exception as e:
         error_msg = str(e)
         traceback_str = traceback.format_exc()
-        sys.stderr.write(f"Error in refresh_leaderboard: {error_msg}\n")
+        sys.stderr.write(f"CRITICAL ERROR in refresh_leaderboard: {error_msg}\n")
         sys.stderr.write(f"Traceback: {traceback_str}\n")
         sys.stderr.flush()
-        raise
+        
+        # Create emergency fallback leaderboard
+        try:
+            sys.stderr.write("Creating emergency fallback leaderboard...\n")
+            sys.stderr.flush()
+            fallback_df = create_fallback_dataframe()
+            return init_leaderboard(fallback_df)
+        except Exception as fallback_error:
+            sys.stderr.write(f"Even fallback failed: {fallback_error}\n")
+            sys.stderr.flush()
+            raise Exception(f"Complete leaderboard failure: {error_msg}")
+
+def create_fallback_dataframe():
+    """Create a minimal valid DataFrame that won't crash the leaderboard"""
+    import pandas as pd
+    import sys
+    
+    sys.stderr.write("Creating fallback DataFrame...\n")
+    sys.stderr.flush()
+    
+    # Create minimal valid data
+    fallback_data = {col: [] for col in COLS}
+    
+    # Add one dummy row to prevent leaderboard component from crashing
+    dummy_row = {}
+    for col in COLS:
+        if col in BENCHMARK_COLS or col == AutoEvalColumn.average.name:
+            dummy_row[col] = 0.0
+        elif col == AutoEvalColumn.model.name:
+            dummy_row[col] = "No models evaluated yet"
+        elif col == AutoEvalColumn.model_type_symbol.name:
+            dummy_row[col] = "?"
+        elif col == AutoEvalColumn.precision.name:
+            dummy_row[col] = "float16"
+        elif col == AutoEvalColumn.model_type.name:
+            dummy_row[col] = "pretrained"
+        elif col == AutoEvalColumn.weight_type.name:
+            dummy_row[col] = "Original"
+        elif col == AutoEvalColumn.architecture.name:
+            dummy_row[col] = "Unknown"
+        elif col == AutoEvalColumn.still_on_hub.name:
+            dummy_row[col] = True
+        elif col == AutoEvalColumn.license.name:
+            dummy_row[col] = "Unknown"
+        elif col == AutoEvalColumn.params.name:
+            dummy_row[col] = 0.0
+        elif col == AutoEvalColumn.likes.name:
+            dummy_row[col] = 0.0
+        elif col == AutoEvalColumn.revision.name:
+            dummy_row[col] = ""
+        else:
+            dummy_row[col] = ""
+    
+    df = pd.DataFrame([dummy_row])
+    sys.stderr.write(f"Fallback DataFrame created with shape: {df.shape}\n")
+    sys.stderr.write(f"Fallback DataFrame columns: {df.columns.tolist()}\n")
+    sys.stderr.flush()
+    
+    return df
 
 def run_perplexity_test(model_name, revision, precision):
     """Run perplexity evaluation on demand."""
@@ -95,7 +186,7 @@ def run_perplexity_test(model_name, revision, precision):
     
     try:
         # Use stderr for more reliable logging in HF Spaces
-        sys.stderr.write(f"\n=== Running Perplexity Test ===\n")
+        sys.stderr.write(f"\n=== RUNNING PERPLEXITY TEST ===\n")
         sys.stderr.write(f"Model: {model_name}\n")
         sys.stderr.write(f"Revision: {revision}\n")
         sys.stderr.write(f"Precision: {precision}\n")
@@ -112,10 +203,16 @@ def run_perplexity_test(model_name, revision, precision):
                 sys.stderr.flush()
                 
                 new_leaderboard = refresh_leaderboard()
-                sys.stderr.write("Leaderboard refresh successful\n")
-                sys.stderr.flush()
                 
-                return f"✅ Perplexity evaluation completed!\nPerplexity: {result:.4f}\n\nResults saved to leaderboard.", new_leaderboard
+                if new_leaderboard is not None:
+                    sys.stderr.write("Leaderboard refresh successful\n")
+                    sys.stderr.flush()
+                    return f"✅ Perplexity evaluation completed!\nPerplexity: {result:.4f}\n\nResults saved and leaderboard updated.", new_leaderboard
+                else:
+                    sys.stderr.write("Leaderboard refresh returned None\n")
+                    sys.stderr.flush()
+                    return f"✅ Perplexity evaluation completed!\nPerplexity: {result:.4f}\n\n⚠️ Results saved but leaderboard update returned None.\n\nPlease refresh the page to see updated results.", None
+                    
             except Exception as refresh_error:
                 # If leaderboard refresh fails, still show success but don't update leaderboard
                 error_msg = str(refresh_error)
@@ -124,7 +221,11 @@ def run_perplexity_test(model_name, revision, precision):
                 sys.stderr.write(f"Traceback: {traceback_str}\n")
                 sys.stderr.flush()
                 
-                return f"✅ Perplexity evaluation completed!\nPerplexity: {result:.4f}\n\n⚠️ Results saved but leaderboard refresh failed: {error_msg}\n\nPlease refresh the page to see updated results.", None
+                # Check if it's the specific "must have a value set" error
+                if "must have a value set" in error_msg.lower():
+                    return f"✅ Perplexity evaluation completed!\nPerplexity: {result:.4f}\n\n⚠️ Results saved but leaderboard component failed to update due to data structure issue.\n\n**Please refresh the page** to see your results in the main leaderboard.", None
+                else:
+                    return f"✅ Perplexity evaluation completed!\nPerplexity: {result:.4f}\n\n⚠️ Results saved but leaderboard refresh failed: {error_msg}\n\nPlease refresh the page to see updated results.", None
         else:
             return f"❌ Evaluation failed: {result}", None
             

leaderboard.py

@@ -0,0 +1,402 @@
+"""gr.Leaderboard() component"""
+
+from __future__ import annotations
+
+import warnings
+from typing import TYPE_CHECKING, Any, Dict, List, Optional, Tuple, Union, Literal
+
+from pandas.api.types import (
+    is_numeric_dtype,
+    is_object_dtype,
+    is_string_dtype,
+    is_bool_dtype,
+)
+import semantic_version
+from dataclasses import dataclass, field
+
+from gradio.components import Component
+from gradio.data_classes import GradioModel
+from gradio.events import Events
+
+if TYPE_CHECKING:
+    import pandas as pd
+    from pandas.io.formats.style import Styler
+
+
+@dataclass
+class SearchColumns:
+    primary_column: str
+    secondary_columns: Optional[List[str]]
+    label: Optional[str] = None
+    placeholder: Optional[str] = None
+
+
+@dataclass
+class SelectColumns:
+    default_selection: Optional[list[str]] = field(default_factory=list)
+    cant_deselect: Optional[list[str]] = field(default_factory=list)
+    allow: bool = True
+    label: Optional[str] = None
+    show_label: bool = True
+    info: Optional[str] = None
+
+
+@dataclass
+class ColumnFilter:
+    column: str
+    type: Literal["slider", "dropdown", "checkboxgroup", "boolean"] = None
+    default: Optional[Union[int, float, List[Tuple[str, str]]]] = None
+    choices: Optional[Union[int, float, List[Tuple[str, str]]]] = None
+    label: Optional[str] = None
+    info: Optional[str] = None
+    show_label: bool = True
+    min: Optional[Union[int, float]] = None
+    max: Optional[Union[int, float]] = None
+
+
+class DataframeData(GradioModel):
+    headers: List[str]
+    data: Union[List[List[Any]], List[Tuple[Any, ...]]]
+    metadata: Optional[Dict[str, Optional[List[Any]]]] = None
+
+
+class Leaderboard(Component):
+    """
+    This component displays a table of value spreadsheet-like component. Can be used to display data as an output component, or as an input to collect data from the user.
+    Demos: filter_records, matrix_transpose, tax_calculator, sort_records
+    """
+
+    EVENTS = [Events.change, Events.input, Events.select]
+
+    data_model = DataframeData
+
+    def __init__(
+        self,
+        value: pd.DataFrame | None = None,
+        *,
+        datatype: str | list[str] = "str",
+        search_columns: list[str] | SearchColumns | None = None,
+        select_columns: list[str] | SelectColumns | None = None,
+        filter_columns: list[str | ColumnFilter] | None = None,
+        bool_checkboxgroup_label: str | None = None,
+        hide_columns: list[str] | None = None,
+        latex_delimiters: list[dict[str, str | bool]] | None = None,
+        label: str | None = None,
+        show_label: bool | None = None,
+        every: float | None = None,
+        height: int = 500,
+        scale: int | None = None,
+        min_width: int = 160,
+        interactive: bool | None = None,
+        visible: bool = True,
+        elem_id: str | None = None,
+        elem_classes: list[str] | str | None = None,
+        render: bool = True,
+        wrap: bool = False,
+        line_breaks: bool = True,
+        column_widths: list[str | int] | None = None,
+    ):
+        """
+        Parameters:
+            value: Default value to display in the DataFrame. Must be a pandas DataFrame.
+            datatype: Datatype of values in sheet. Can be provided per column as a list of strings, or for the entire sheet as a single string. Valid datatypes are "str", "number", "bool", "date", and "markdown".
+            search_columns: See Configuration section of docs for details.
+            select_columns: See Configuration section of docs for details.
+            filter_columns: See Configuration section of docs for details.
+            bool_checkboxgroup_label: Label for the checkboxgroup filter for boolean columns.
+            hide_columns: List of columns to hide by default. They will not be displayed in the table but they can still be used for searching, filtering.
+            label: The label for this component. Appears above the component and is also used as the header if there are a table of examples for this component. If None and used in a `gr.Interface`, the label will be the name of the parameter this component is assigned to.
+            latex_delimiters: A list of dicts of the form {"left": open delimiter (str), "right": close delimiter (str), "display": whether to display in newline (bool)} that will be used to render LaTeX expressions. If not provided, `latex_delimiters` is set to `[{ "left": "$$", "right": "$$", "display": True }]`, so only expressions enclosed in $$ delimiters will be rendered as LaTeX, and in a new line. Pass in an empty list to disable LaTeX rendering. For more information, see the [KaTeX documentation](https://katex.org/docs/autorender.html). Only applies to columns whose datatype is "markdown".
+            label: The label for this component. Appears above the component and is also used as the header if there are a table of examples for this component. If None and used in a `gr.Interface`, the label will be the name of the parameter this component is assigned to.
+            show_label: if True, will display label.
+            every: If `value` is a callable, run the function 'every' number of seconds while the client connection is open. Has no effect otherwise. The event can be accessed (e.g. to cancel it) via this component's .load_event attribute.
+            height: The maximum height of the dataframe, specified in pixels if a number is passed, or in CSS units if a string is passed. If more rows are created than can fit in the height, a scrollbar will appear.
+            scale: relative size compared to adjacent Components. For example if Components A and B are in a Row, and A has scale=2, and B has scale=1, A will be twice as wide as B. Should be an integer. scale applies in Rows, and to top-level Components in Blocks where fill_height=True.
+            min_width: minimum pixel width, will wrap if not sufficient screen space to satisfy this value. If a certain scale value results in this Component being narrower than min_width, the min_width parameter will be respected first.
+            interactive: if True, will allow users to edit the dataframe; if False, can only be used to display data. If not provided, this is inferred based on whether the component is used as an input or output.
+            visible: If False, component will be hidden.
+            elem_id: An optional string that is assigned as the id of this component in the HTML DOM. Can be used for targeting CSS styles.
+            elem_classes: An optional list of strings that are assigned as the classes of this component in the HTML DOM. Can be used for targeting CSS styles.
+            render: If False, component will not render be rendered in the Blocks context. Should be used if the intention is to assign event listeners now but render the component later.
+            wrap: If True, the text in table cells will wrap when appropriate. If False and the `column_width` parameter is not set, the column widths will expand based on the cell contents and the table may need to be horizontally scrolled. If `column_width` is set, then any overflow text will be hidden.
+            line_breaks: If True (default), will enable Github-flavored Markdown line breaks in chatbot messages. If False, single new lines will be ignored. Only applies for columns of type "markdown."
+            column_widths: An optional list representing the width of each column. The elements of the list should be in the format "100px" (ints are also accepted and converted to pixel values) or "10%". If not provided, the column widths will be automatically determined based on the content of the cells. Setting this parameter will cause the browser to try to fit the table within the page width.
+        """
+        if value is None:
+            raise ValueError("Leaderboard component must have a value set.")
+        self.wrap = wrap
+        self.headers = [str(s) for s in value.columns]
+        self.datatype = datatype
+        self.search_columns = self._get_search_columns(search_columns)
+        self.bool_checkboxgroup_label = bool_checkboxgroup_label
+        self.select_columns_config = self._get_select_columns(select_columns, value)
+        self.filter_columns = self._get_column_filter_configs(filter_columns, value)
+        self.raise_error_if_incorrect_config()
+
+        self.hide_columns = hide_columns or []
+        self.col_count = (len(self.headers), "fixed")
+        self.row_count = (value.shape[0], "fixed")
+
+        if latex_delimiters is None:
+            latex_delimiters = [{"left": "$$", "right": "$$", "display": True}]
+        self.latex_delimiters = latex_delimiters
+        self.height = height
+        self.line_breaks = line_breaks
+        self.column_widths = [
+            w if isinstance(w, str) else f"{w}px" for w in (column_widths or [])
+        ]
+        super().__init__(
+            label=label,
+            every=every,
+            show_label=show_label,
+            scale=scale,
+            min_width=min_width,
+            interactive=interactive,
+            visible=visible,
+            elem_id=elem_id,
+            elem_classes=elem_classes,
+            render=render,
+            value=value,
+        )
+    
+    def raise_error_if_incorrect_config(self):
+        for col in [self.search_columns.primary_column, *self.search_columns.secondary_columns]:
+            if col not in self.headers:
+                raise ValueError(f"Column '{col}' not found in the DataFrame headers.")
+        for col in self.select_columns_config.default_selection + self.select_columns_config.cant_deselect:
+            if col not in self.headers:
+                raise ValueError(f"Column '{col}' not found in the DataFrame headers.")
+        for col in [col.column for col in self.filter_columns]:
+            if col not in self.headers:
+                raise ValueError(f"Column '{col}' not found in the DataFrame headers.")
+
+    @staticmethod
+    def _get_best_filter_type(
+        column: str, value: pd.DataFrame
+    ) -> Literal["slider", "checkboxgroup", "dropdown", "checkbox"]:
+        if is_bool_dtype(value[column]):
+            return "checkbox"
+        if is_numeric_dtype(value[column]):
+            return "slider"
+        if is_string_dtype(value[column]) or is_object_dtype(value[column]):
+            return "checkboxgroup"
+        warnings.warn(
+            f"{column}'s type is not numeric or string, defaulting to checkboxgroup filter type.",
+            UserWarning,
+        )
+        return "checkboxgroup"
+
+    @staticmethod
+    def _get_column_filter_configs(
+        columns: list[str | ColumnFilter] | None, value: pd.DataFrame
+    ) -> list[ColumnFilter]:
+        if columns is None:
+            return []
+        if not isinstance(columns, list):
+            raise ValueError(
+                "Columns must be a list of strings or ColumnFilter objects"
+            )
+        return [
+            Leaderboard._get_column_filter_config(column, value) for column in columns
+        ]
+
+    @staticmethod
+    def _get_column_filter_config(column: str | ColumnFilter, value: pd.DataFrame):
+        column_name = column if isinstance(column, str) else column.column
+        best_filter_type = Leaderboard._get_best_filter_type(column_name, value)
+        min_val = None
+        max_val = None
+        if best_filter_type == "slider":
+            default = [
+                value[column_name].quantile(0.25),
+                value[column_name].quantile(0.70),
+            ]
+            min_val = value[column_name].min()
+            max_val = value[column_name].max()
+            choices = None
+        elif best_filter_type == "checkbox":
+            default = False
+            choices = None
+        else:
+            default = value[column_name].unique().tolist()
+            default = [(s, s) for s in default]
+            choices = default
+        if isinstance(column, ColumnFilter):
+            if column.type == "boolean":
+                column.type = "checkbox"
+            if not column.type:
+                column.type = best_filter_type
+            if column.default is None:
+                column.default = default
+            if not column.choices:
+                column.choices = choices
+            if min_val is not None and max_val is not None:
+                column.min = min_val
+                column.max = max_val
+            return column
+        if isinstance(column, str):
+            return ColumnFilter(
+                column=column,
+                type=best_filter_type,
+                default=default,
+                choices=choices,
+                min=min_val,
+                max=max_val,
+            )
+        raise ValueError(f"Columns {column} must be a string or a ColumnFilter object")
+
+    @staticmethod
+    def _get_search_columns(
+        search_columns: list[str] | SearchColumns | None,
+    ) -> SearchColumns:
+        if search_columns is None:
+            return SearchColumns(primary_column=None, secondary_columns=[])
+        if isinstance(search_columns, SearchColumns):
+            return search_columns
+        if isinstance(search_columns, list):
+            return SearchColumns(
+                primary_column=search_columns[0], secondary_columns=search_columns[1:]
+            )
+        raise ValueError(
+            "search_columns must be a list of strings or a SearchColumns object"
+        )
+
+    @staticmethod
+    def _get_select_columns(
+        select_columns: list[str] | SelectColumns | None,
+        value: pd.DataFrame,
+    ) -> SelectColumns:
+        if select_columns is None:
+            return SelectColumns(allow=False)
+        if isinstance(select_columns, SelectColumns):
+            if not select_columns.default_selection:
+                select_columns.default_selection = value.columns.tolist()
+            return select_columns
+        if isinstance(select_columns, list):
+            return SelectColumns(default_selection=select_columns, allow=True)
+        raise ValueError(
+            "select_columns must be a list of strings or a SelectColumns object"
+        )
+
+    def get_config(self):
+        return {
+            "row_count": self.row_count,
+            "col_count": self.col_count,
+            "headers": self.headers,
+            "select_columns_config": self.select_columns_config,
+            **super().get_config(),
+        }
+
+    def preprocess(self, payload: DataframeData) -> pd.DataFrame:
+        """
+        Parameters:
+            payload: the uploaded spreadsheet data as an object with `headers` and `data` attributes
+        Returns:
+            Passes the uploaded spreadsheet data as a `pandas.DataFrame`, `numpy.array`, `polars.DataFrame`, or native 2D Python `list[list]` depending on `type`
+        """
+        import pandas as pd
+
+        if payload.headers is not None:
+            return pd.DataFrame(
+                [] if payload.data == [[]] else payload.data,
+                columns=payload.headers,
+            )
+        else:
+            return pd.DataFrame(payload.data)
+
+    def postprocess(self, value: pd.DataFrame) -> DataframeData:
+        """
+        Parameters:
+            value: Expects data any of these formats: `pandas.DataFrame`, `pandas.Styler`, `numpy.array`, `polars.DataFrame`, `list[list]`, `list`, or a `dict` with keys 'data' (and optionally 'headers'), or `str` path to a csv, which is rendered as the spreadsheet.
+        Returns:
+            the uploaded spreadsheet data as an object with `headers` and `data` attributes
+        """
+        import pandas as pd
+        from pandas.io.formats.style import Styler
+
+        if value is None:
+            return self.postprocess(pd.DataFrame({"column 1": []}))
+        if isinstance(value, (str, pd.DataFrame)):
+            if isinstance(value, str):
+                value = pd.read_csv(value)  # type: ignore
+            if len(value) == 0:
+                return DataframeData(
+                    headers=list(value.columns),  # type: ignore
+                    data=[[]],  # type: ignore
+                )
+            return DataframeData(
+                headers=list(value.columns),  # type: ignore
+                data=value.to_dict(orient="split")["data"],  # type: ignore
+            )
+        elif isinstance(value, Styler):
+            if semantic_version.Version(pd.__version__) < semantic_version.Version(
+                "1.5.0"
+            ):
+                raise ValueError(
+                    "Styler objects are only supported in pandas version 1.5.0 or higher. Please try: `pip install --upgrade pandas` to use this feature."
+                )
+            if self.interactive:
+                warnings.warn(
+                    "Cannot display Styler object in interactive mode. Will display as a regular pandas dataframe instead."
+                )
+            df: pd.DataFrame = value.data  # type: ignore
+            if len(df) == 0:
+                return DataframeData(
+                    headers=list(df.columns),
+                    data=[[]],
+                    metadata=self.__extract_metadata(value),  # type: ignore
+                )
+            return DataframeData(
+                headers=list(df.columns),
+                data=df.to_dict(orient="split")["data"],  # type: ignore
+                metadata=self.__extract_metadata(value),  # type: ignore
+            )
+
+    @staticmethod
+    def __get_cell_style(cell_id: str, cell_styles: list[dict]) -> str:
+        styles_for_cell = []
+        for style in cell_styles:
+            if cell_id in style.get("selectors", []):
+                styles_for_cell.extend(style.get("props", []))
+        styles_str = "; ".join([f"{prop}: {value}" for prop, value in styles_for_cell])
+        return styles_str
+
+    @staticmethod
+    def __extract_metadata(df: Styler) -> dict[str, list[list]]:
+        metadata = {"display_value": [], "styling": []}
+        style_data = df._compute()._translate(None, None)  # type: ignore
+        cell_styles = style_data.get("cellstyle", [])
+        for i in range(len(style_data["body"])):
+            metadata["display_value"].append([])
+            metadata["styling"].append([])
+            for j in range(len(style_data["body"][i])):
+                cell_type = style_data["body"][i][j]["type"]
+                if cell_type != "td":
+                    continue
+                display_value = style_data["body"][i][j]["display_value"]
+                cell_id = style_data["body"][i][j]["id"]
+                styles_str = Leaderboard.__get_cell_style(cell_id, cell_styles)
+                metadata["display_value"][i].append(display_value)
+                metadata["styling"][i].append(styles_str)
+        return metadata
+
+    def process_example(
+        self,
+        value: pd.DataFrame | Styler | str | None,
+    ):
+        import pandas as pd
+
+        if value is None:
+            return ""
+        value_df_data = self.postprocess(value)
+        value_df = pd.DataFrame(value_df_data.data, columns=value_df_data.headers)
+        return value_df.head(n=5).to_dict(orient="split")["data"]
+
+    def example_payload(self) -> Any:
+        return {"headers": ["a", "b"], "data": [["foo", "bar"]]}
+
+    def example_inputs(self) -> Any:
+        return self.example_value()
+
+    def example_value(self) -> Any:
+        return {"headers": ["a", "b"], "data": [["foo", "bar"]]}

logs.txt

@@ -0,0 +1,266 @@
+==== Application Startup at 2025-07-25 22:55:49 =====
+
+
+.gitattributes:   0%|          | 0.00/2.46k [00:00<?, ?B/s]
+.gitattributes: 100%|██████████| 2.46k/2.46k [00:00<00:00, 10.5MB/s]
+
+(…)enai-community_gpt2_20250725_231201.json:   0%|          | 0.00/209 [00:00<?, ?B/s]
+(…)enai-community_gpt2_20250725_231201.json: 100%|██████████| 209/209 [00:00<00:00, 1.71MB/s]
+
+(…)enai-community_gpt2_20250725_233155.json:   0%|          | 0.00/209 [00:00<?, ?B/s]
+(…)enai-community_gpt2_20250725_233155.json: 100%|██████████| 209/209 [00:00<00:00, 1.26MB/s]
+
+(…)enai-community_gpt2_20250725_235115.json:   0%|          | 0.00/209 [00:00<?, ?B/s]
+(…)enai-community_gpt2_20250725_235115.json: 100%|██████████| 209/209 [00:00<00:00, 2.02MB/s]
+
+(…)enai-community_gpt2_20250725_235748.json:   0%|          | 0.00/209 [00:00<?, ?B/s]
+(…)enai-community_gpt2_20250725_235748.json: 100%|██████████| 209/209 [00:00<00:00, 2.08MB/s]
+
+(…)enai-community_gpt2_20250726_000358.json:   0%|          | 0.00/209 [00:00<?, ?B/s]
+(…)enai-community_gpt2_20250726_000358.json: 100%|██████████| 209/209 [00:00<00:00, 1.54MB/s]
+
+(…)enai-community_gpt2_20250726_000650.json:   0%|          | 0.00/209 [00:00<?, ?B/s]
+(…)enai-community_gpt2_20250726_000650.json: 100%|██████████| 209/209 [00:00<00:00, 2.35MB/s]
+
+=== Starting leaderboard creation ===
+Looking for results in: ./eval-results
+Expected columns: ['T', 'Model', 'Average ⬆️', 'Perplexity', 'Type', 'Architecture', 'Precision', 'Hub License', '#Params (B)', 'Hub ❤️', 'Available on the hub', 'Model sha']
+Benchmark columns: ['Perplexity']
+
+Searching for result files in: ./eval-results
+Found 6 result files
+
+Processing file: ./eval-results/openai-community/results_openai-community_gpt2_20250725_231201.json
+
+config.json:   0%|          | 0.00/665 [00:00<?, ?B/s]
+config.json: 100%|██████████| 665/665 [00:00<00:00, 6.14MB/s]
+Created result object for: openai-community/gpt2
+Added new result for openai-community_gpt2_float16
+
+Processing file: ./eval-results/openai-community/results_openai-community_gpt2_20250725_233155.json
+Created result object for: openai-community/gpt2
+Updated existing result for openai-community_gpt2_float16
+
+Processing file: ./eval-results/openai-community/results_openai-community_gpt2_20250725_235115.json
+Created result object for: openai-community/gpt2
+Updated existing result for openai-community_gpt2_float16
+
+Processing file: ./eval-results/openai-community/results_openai-community_gpt2_20250725_235748.json
+Created result object for: openai-community/gpt2
+Updated existing result for openai-community_gpt2_float16
+
+Processing file: ./eval-results/openai-community/results_openai-community_gpt2_20250726_000358.json
+Created result object for: openai-community/gpt2
+Updated existing result for openai-community_gpt2_float16
+
+Processing file: ./eval-results/openai-community/results_openai-community_gpt2_20250726_000650.json
+Created result object for: openai-community/gpt2
+Updated existing result for openai-community_gpt2_float16
+
+Processing 1 evaluation results
+
+Converting result to dict for: openai-community/gpt2
+
+Processing result for model: openai-community/gpt2
+Raw results: {'perplexity': 20.663532257080078}
+Calculated average score: 69.7162958010531
+Added perplexity score 20.663532257080078 under column Perplexity
+Final data dict keys: ['eval_name', 'Precision', 'Type', 'T', 'Weight type', 'Architecture', 'Model', 'Model sha', 'Average ⬆️', 'Available on the hub', 'Hub License', '#Params (B)', 'Hub ❤️', 'Perplexity']
+Successfully converted and added result
+
+Returning 1 processed results
+
+Found 1 raw results
+
+Processing result for model: openai-community/gpt2
+Raw results: {'perplexity': 20.663532257080078}
+Calculated average score: 69.7162958010531
+Added perplexity score 20.663532257080078 under column Perplexity
+Final data dict keys: ['eval_name', 'Precision', 'Type', 'T', 'Weight type', 'Architecture', 'Model', 'Model sha', 'Average ⬆️', 'Available on the hub', 'Hub License', '#Params (B)', 'Hub ❤️', 'Perplexity']
+Successfully processed result 1/1: openai-community/gpt2
+
+Converted to 1 JSON records
+Sample record keys: ['eval_name', 'Precision', 'Type', 'T', 'Weight type', 'Architecture', 'Model', 'Model sha', 'Average ⬆️', 'Available on the hub', 'Hub License', '#Params (B)', 'Hub ❤️', 'Perplexity']
+
+Created DataFrame with columns: ['eval_name', 'Precision', 'Type', 'T', 'Weight type', 'Architecture', 'Model', 'Model sha', 'Average ⬆️', 'Available on the hub', 'Hub License', '#Params (B)', 'Hub ❤️', 'Perplexity']
+DataFrame shape: (1, 14)
+
+Sorted DataFrame by average
+
+Selected and rounded columns
+
+Final DataFrame shape after filtering: (1, 12)
+Final columns: ['T', 'Model', 'Average ⬆️', 'Perplexity', 'Type', 'Architecture', 'Precision', 'Hub License', '#Params (B)', 'Hub ❤️', 'Available on the hub', 'Model sha']
+
+=== Initializing Leaderboard ===
+DataFrame shape: (1, 12)
+DataFrame columns: ['T', 'Model', 'Average ⬆️', 'Perplexity', 'Type', 'Architecture', 'Precision', 'Hub License', '#Params (B)', 'Hub ❤️', 'Available on the hub', 'Model sha']
+* Running on local URL:  http://0.0.0.0:7860, with SSR ⚡ (experimental, to disable set `ssr=False` in `launch()`)
+
+To create a public link, set `share=True` in `launch()`.
+
+=== Running Perplexity Test ===
+Model: EleutherAI/gpt-neo-1.3B
+Revision: main
+Precision: float16
+Starting dynamic evaluation for EleutherAI/gpt-neo-1.3B
+Running perplexity evaluation...
+Loading model: EleutherAI/gpt-neo-1.3B (revision: main)
+Loading tokenizer...
+
+tokenizer_config.json:   0%|          | 0.00/200 [00:00<?, ?B/s]
+tokenizer_config.json: 100%|██████████| 200/200 [00:00<00:00, 1.64MB/s]
+
+config.json:   0%|          | 0.00/1.35k [00:00<?, ?B/s]
+config.json: 100%|██████████| 1.35k/1.35k [00:00<00:00, 9.77MB/s]
+
+vocab.json:   0%|          | 0.00/798k [00:00<?, ?B/s]
+vocab.json: 100%|██████████| 798k/798k [00:00<00:00, 27.9MB/s]
+
+merges.txt:   0%|          | 0.00/456k [00:00<?, ?B/s]
+merges.txt: 100%|██████████| 456k/456k [00:00<00:00, 3.54MB/s]
+
+special_tokens_map.json:   0%|          | 0.00/90.0 [00:00<?, ?B/s]
+special_tokens_map.json: 100%|██████████| 90.0/90.0 [00:00<00:00, 1.05MB/s]
+Tokenizer loaded successfully
+Loading model...
+
+model.safetensors:   0%|          | 0.00/5.31G [00:00<?, ?B/s]
+model.safetensors:   0%|          | 778k/5.31G [00:01<2:15:00, 656kB/s]
+model.safetensors:   0%|          | 7.69M/5.31G [00:02<23:51, 3.70MB/s]
+model.safetensors:   1%|▏         | 74.7M/5.31G [00:03<03:29, 25.0MB/s]
+model.safetensors:   9%|▉         | 496M/5.31G [00:04<00:31, 153MB/s]  
+model.safetensors:  19%|█▉        | 1.03G/5.31G [00:06<00:16, 263MB/s]
+model.safetensors:  25%|██▍       | 1.32G/5.31G [00:07<00:16, 235MB/s]
+model.safetensors:  38%|███▊      | 1.99G/5.31G [00:08<00:09, 346MB/s]
+model.safetensors:  47%|████▋     | 2.51G/5.31G [00:09<00:07, 379MB/s]
+model.safetensors:  59%|█████▊    | 3.11G/5.31G [00:10<00:05, 429MB/s]
+model.safetensors:  69%|██████▊   | 3.65G/5.31G [00:11<00:03, 451MB/s]
+model.safetensors:  80%|███████▉  | 4.24G/5.31G [00:13<00:02, 477MB/s]
+model.safetensors:  91%|█████████ | 4.84G/5.31G [00:14<00:00, 494MB/s]
+model.safetensors: 100%|██████████| 5.31G/5.31G [00:14<00:00, 355MB/s]
+Model loaded successfully
+Tokenizing input text...
+Tokenized input shape: torch.Size([1, 141])
+Moved inputs to device: cpu
+Running forward pass...
+Calculated loss: 1.78515625
+Final perplexity: 5.9609375
+Perplexity evaluation completed: 5.9609375
+Created result structure: {'config': {'model_dtype': 'torch.float16', 'model_name': 'EleutherAI/gpt-neo-1.3B', 'model_sha': 'main'}, 'results': {'perplexity': {'perplexity': 5.9609375}}}
+Saving result to: ./eval-results/EleutherAI/results_EleutherAI_gpt-neo-1.3B_20250726_010247.json
+Result file saved locally
+Uploading to HF dataset: ahmedsqrd/results
+Upload completed successfully
+Evaluation result - Success: True, Result: 5.9609375
+Attempting to refresh leaderboard...
+Refreshing leaderboard data...
+
+=== Starting leaderboard creation ===
+Looking for results in: ./eval-results
+Expected columns: ['T', 'Model', 'Average ⬆️', 'Perplexity', 'Type', 'Architecture', 'Precision', 'Hub License', '#Params (B)', 'Hub ❤️', 'Available on the hub', 'Model sha']
+Benchmark columns: ['Perplexity']
+
+Searching for result files in: ./eval-results
+Found 7 result files
+
+Processing file: ./eval-results/openai-community/results_openai-community_gpt2_20250725_231201.json
+Created result object for: openai-community/gpt2
+Added new result for openai-community_gpt2_float16
+
+Processing file: ./eval-results/openai-community/results_openai-community_gpt2_20250725_233155.json
+Created result object for: openai-community/gpt2
+Updated existing result for openai-community_gpt2_float16
+
+Processing file: ./eval-results/openai-community/results_openai-community_gpt2_20250725_235115.json
+Created result object for: openai-community/gpt2
+Updated existing result for openai-community_gpt2_float16
+
+Processing file: ./eval-results/openai-community/results_openai-community_gpt2_20250725_235748.json
+Created result object for: openai-community/gpt2
+Updated existing result for openai-community_gpt2_float16
+
+Processing file: ./eval-results/openai-community/results_openai-community_gpt2_20250726_000358.json
+Created result object for: openai-community/gpt2
+Updated existing result for openai-community_gpt2_float16
+
+Processing file: ./eval-results/openai-community/results_openai-community_gpt2_20250726_000650.json
+Created result object for: openai-community/gpt2
+Updated existing result for openai-community_gpt2_float16
+
+Processing file: ./eval-results/EleutherAI/results_EleutherAI_gpt-neo-1.3B_20250726_010247.json
+Created result object for: EleutherAI/gpt-neo-1.3B
+Added new result for EleutherAI_gpt-neo-1.3B_float16
+
+Processing 2 evaluation results
+
+Converting result to dict for: openai-community/gpt2
+
+Processing result for model: openai-community/gpt2
+Raw results: {'perplexity': 20.663532257080078}
+Calculated average score: 69.7162958010531
+Added perplexity score 20.663532257080078 under column Perplexity
+Final data dict keys: ['eval_name', 'Precision', 'Type', 'T', 'Weight type', 'Architecture', 'Model', 'Model sha', 'Average ⬆️', 'Available on the hub', 'Hub License', '#Params (B)', 'Hub ❤️', 'Perplexity']
+Successfully converted and added result
+
+Converting result to dict for: EleutherAI/gpt-neo-1.3B
+
+Processing result for model: EleutherAI/gpt-neo-1.3B
+Raw results: {'perplexity': 5.9609375}
+Calculated average score: 82.1477223263516
+Added perplexity score 5.9609375 under column Perplexity
+Final data dict keys: ['eval_name', 'Precision', 'Type', 'T', 'Weight type', 'Architecture', 'Model', 'Model sha', 'Average ⬆️', 'Available on the hub', 'Hub License', '#Params (B)', 'Hub ❤️', 'Perplexity']
+Successfully converted and added result
+
+Returning 2 processed results
+
+Found 2 raw results
+
+Processing result for model: openai-community/gpt2
+Raw results: {'perplexity': 20.663532257080078}
+Calculated average score: 69.7162958010531
+Added perplexity score 20.663532257080078 under column Perplexity
+Final data dict keys: ['eval_name', 'Precision', 'Type', 'T', 'Weight type', 'Architecture', 'Model', 'Model sha', 'Average ⬆️', 'Available on the hub', 'Hub License', '#Params (B)', 'Hub ❤️', 'Perplexity']
+Successfully processed result 1/2: openai-community/gpt2
+
+Processing result for model: EleutherAI/gpt-neo-1.3B
+Raw results: {'perplexity': 5.9609375}
+Calculated average score: 82.1477223263516
+Added perplexity score 5.9609375 under column Perplexity
+Final data dict keys: ['eval_name', 'Precision', 'Type', 'T', 'Weight type', 'Architecture', 'Model', 'Model sha', 'Average ⬆️', 'Available on the hub', 'Hub License', '#Params (B)', 'Hub ❤️', 'Perplexity']
+Successfully processed result 2/2: EleutherAI/gpt-neo-1.3B
+
+Converted to 2 JSON records
+Sample record keys: ['eval_name', 'Precision', 'Type', 'T', 'Weight type', 'Architecture', 'Model', 'Model sha', 'Average ⬆️', 'Available on the hub', 'Hub License', '#Params (B)', 'Hub ❤️', 'Perplexity']
+
+Created DataFrame with columns: ['eval_name', 'Precision', 'Type', 'T', 'Weight type', 'Architecture', 'Model', 'Model sha', 'Average ⬆️', 'Available on the hub', 'Hub License', '#Params (B)', 'Hub ❤️', 'Perplexity']
+DataFrame shape: (2, 14)
+
+Sorted DataFrame by average
+
+Selected and rounded columns
+
+Final DataFrame shape after filtering: (2, 12)
+Final columns: ['T', 'Model', 'Average ⬆️', 'Perplexity', 'Type', 'Architecture', 'Precision', 'Hub License', '#Params (B)', 'Hub ❤️', 'Available on the hub', 'Model sha']
+Got DataFrame with shape: (2, 12)
+DataFrame columns: ['T', 'Model', 'Average ⬆️', 'Perplexity', 'Type', 'Architecture', 'Precision', 'Hub License', '#Params (B)', 'Hub ❤️', 'Available on the hub', 'Model sha']
+Creating leaderboard with valid DataFrame
+
+=== Initializing Leaderboard ===
+DataFrame shape: (2, 12)
+DataFrame columns: ['T', 'Model', 'Average ⬆️', 'Perplexity', 'Type', 'Architecture', 'Precision', 'Hub License', '#Params (B)', 'Hub ❤️', 'Available on the hub', 'Model sha']
+Leaderboard refresh successful
+Traceback (most recent call last):
+  File "/usr/local/lib/python3.10/site-packages/gradio/queueing.py", line 625, in process_events
+    response = await route_utils.call_process_api(
+  File "/usr/local/lib/python3.10/site-packages/gradio/route_utils.py", line 322, in call_process_api
+    output = await app.get_blocks().process_api(
+  File "/usr/local/lib/python3.10/site-packages/gradio/blocks.py", line 2106, in process_api
+    data = await self.postprocess_data(block_fn, result["prediction"], state)
+  File "/usr/local/lib/python3.10/site-packages/gradio/blocks.py", line 1899, in postprocess_data
+    state[block._id] = block.__class__(**kwargs)
+  File "/usr/local/lib/python3.10/site-packages/gradio/component_meta.py", line 181, in wrapper
+    return fn(self, **kwargs)
+  File "/usr/local/lib/python3.10/site-packages/gradio_leaderboard/leaderboard.py", line 126, in __init__
+    raise ValueError("Leaderboard component must have a value set.")
+ValueError: Leaderboard component must have a value set.

src/display/utils.py

@@ -1,5 +1,6 @@
 from dataclasses import dataclass, make_dataclass
 from enum import Enum
+import sys
 
 import pandas as pd
 
@@ -29,7 +30,10 @@ auto_eval_column_dict.append(["model", ColumnContent, ColumnContent("Model", "ma
 auto_eval_column_dict.append(["average", ColumnContent, ColumnContent("Average ⬆️", "number", True)])
 for task in Tasks:
     # Use exact column name from Tasks
-    auto_eval_column_dict.append([task.name, ColumnContent, ColumnContent(task.value.col_name, "number", True)])
+    task_col_name = task.value.col_name
+    sys.stderr.write(f"Adding task column: {task.name} -> column name: {task_col_name}\n")
+    auto_eval_column_dict.append([task.name, ColumnContent, ColumnContent(task_col_name, "number", True)])
+    sys.stderr.flush()
 # Model information
 auto_eval_column_dict.append(["model_type", ColumnContent, ColumnContent("Type", "str", False)])
 auto_eval_column_dict.append(["architecture", ColumnContent, ColumnContent("Architecture", "str", False)])
@@ -44,6 +48,13 @@ auto_eval_column_dict.append(["revision", ColumnContent, ColumnContent("Model sh
 # We use make dataclass to dynamically fill the scores from Tasks
 AutoEvalColumn = make_dataclass("AutoEvalColumn", auto_eval_column_dict, frozen=True)
 
+# Debug: Print the created columns
+sys.stderr.write("\n=== CREATED AUTOEVALCOLUMN ===\n")
+for field_obj in fields(AutoEvalColumn):
+    sys.stderr.write(f"Field: {field_obj.name} -> Display: {field_obj.name}\n")
+sys.stderr.write("=== END AUTOEVALCOLUMN ===\n")
+sys.stderr.flush()
+
 ## For the queue columns in the submission tab
 @dataclass(frozen=True)
 class EvalQueueColumn:  # Queue column
@@ -103,9 +114,14 @@ class Precision(Enum):
 
 # Column selection
 COLS = [c.name for c in fields(AutoEvalColumn) if not c.hidden]
+sys.stderr.write(f"\n=== FINAL COLUMN SETUP ===\n")
+sys.stderr.write(f"COLS: {COLS}\n")
 
 EVAL_COLS = [c.name for c in fields(EvalQueueColumn)]
 EVAL_TYPES = [c.type for c in fields(EvalQueueColumn)]
 
 BENCHMARK_COLS = [t.value.col_name for t in Tasks]
+sys.stderr.write(f"BENCHMARK_COLS: {BENCHMARK_COLS}\n")
+sys.stderr.write(f"=== END COLUMN SETUP ===\n")
+sys.stderr.flush()
 

src/leaderboard/read_evals.py

@@ -78,56 +78,78 @@ class EvalResult:
         """Converts the Eval Result to a dict compatible with our dataframe display"""
         import sys
         
-        sys.stderr.write(f"\nProcessing result for model: {self.full_model}\n")
+        sys.stderr.write(f"\n=== PROCESSING RESULT TO_DICT ===\n")
+        sys.stderr.write(f"Processing result for model: {self.full_model}\n")
         sys.stderr.write(f"Raw results: {self.results}\n")
+        sys.stderr.write(f"Model precision: {self.precision}\n")
+        sys.stderr.write(f"Model type: {self.model_type}\n")
+        sys.stderr.write(f"Weight type: {self.weight_type}\n")
         sys.stderr.flush()
         
         # Calculate average, handling perplexity (lower is better)
         scores = []
         perplexity_score = None
+        sys.stderr.write(f"Available tasks: {[task.name for task in Tasks]}\n")
+        
         for task in Tasks:
+            sys.stderr.write(f"Looking for task: {task.value.benchmark} in results\n")
             if task.value.benchmark in self.results:
                 score = self.results[task.value.benchmark]
                 perplexity_score = score  # Save the raw score
+                sys.stderr.write(f"Found score for {task.value.benchmark}: {score}\n")
                 # Convert perplexity to a 0-100 scale where lower perplexity = higher score
                 # Using a log scale since perplexity can vary widely
                 # Cap at 100 for very low perplexity and 0 for very high perplexity
                 score = max(0, min(100, 100 * (1 - math.log(score) / 10)))
                 scores.append(score)
+                sys.stderr.write(f"Converted score: {score}\n")
+            else:
+                sys.stderr.write(f"Task {task.value.benchmark} not found in results\n")
+            sys.stderr.flush()
         
         average = sum(scores) / len(scores) if scores else 0
         sys.stderr.write(f"Calculated average score: {average}\n")
         sys.stderr.flush()
         
-        data_dict = {
-            "eval_name": self.eval_name,  # not a column, just a save name,
-            AutoEvalColumn.precision.name: self.precision.value.name,
-            AutoEvalColumn.model_type.name: self.model_type.value.name,
-            AutoEvalColumn.model_type_symbol.name: self.model_type.value.symbol,
-            AutoEvalColumn.weight_type.name: self.weight_type.value.name,
-            AutoEvalColumn.architecture.name: self.architecture,
-            AutoEvalColumn.model.name: make_clickable_model(self.full_model),
-            AutoEvalColumn.revision.name: self.revision,
-            AutoEvalColumn.average.name: average,
-            AutoEvalColumn.still_on_hub.name: self.still_on_hub,
-            # Add missing columns with default values
-            AutoEvalColumn.license.name: "Unknown",  # Default license
-            AutoEvalColumn.params.name: 0,  # Default params
-            AutoEvalColumn.likes.name: 0,  # Default likes
-        }
-
-        # Add perplexity score with the exact column name from Tasks
-        if perplexity_score is not None:
-            data_dict[Tasks.task0.value.col_name] = perplexity_score
-            sys.stderr.write(f"Added perplexity score {perplexity_score} under column {Tasks.task0.value.col_name}\n")
-            sys.stderr.flush()
-        else:
-            data_dict[Tasks.task0.value.col_name] = None
-            sys.stderr.write(f"No perplexity score found for column {Tasks.task0.value.col_name}\n")
+        # Create data dictionary with comprehensive debugging
+        data_dict = {}
+        
+        # Add core columns
+        data_dict["eval_name"] = self.eval_name
+        data_dict[AutoEvalColumn.precision.name] = self.precision.value.name
+        data_dict[AutoEvalColumn.model_type.name] = self.model_type.value.name
+        data_dict[AutoEvalColumn.model_type_symbol.name] = self.model_type.value.symbol
+        data_dict[AutoEvalColumn.weight_type.name] = self.weight_type.value.name
+        data_dict[AutoEvalColumn.architecture.name] = self.architecture
+        data_dict[AutoEvalColumn.model.name] = make_clickable_model(self.full_model)
+        data_dict[AutoEvalColumn.revision.name] = self.revision
+        data_dict[AutoEvalColumn.average.name] = average
+        data_dict[AutoEvalColumn.still_on_hub.name] = self.still_on_hub
+        
+        # Add default values for missing model info
+        data_dict[AutoEvalColumn.license.name] = "Unknown"
+        data_dict[AutoEvalColumn.params.name] = 0
+        data_dict[AutoEvalColumn.likes.name] = 0
+        
+        sys.stderr.write(f"Created base data_dict with {len(data_dict)} columns\n")
+        sys.stderr.flush()
+
+        # Add task-specific scores
+        for task in Tasks:
+            task_col_name = task.value.col_name
+            if task.value.benchmark in self.results:
+                task_score = self.results[task.value.benchmark]
+                data_dict[task_col_name] = task_score
+                sys.stderr.write(f"Added task score: {task_col_name} = {task_score}\n")
+            else:
+                data_dict[task_col_name] = None
+                sys.stderr.write(f"Added None for missing task: {task_col_name}\n")
             sys.stderr.flush()
 
-        sys.stderr.write(f"Final data dict keys: {list(data_dict.keys())}\n")
+        sys.stderr.write(f"Final data dict has {len(data_dict)} columns: {list(data_dict.keys())}\n")
+        sys.stderr.write(f"=== END PROCESSING RESULT TO_DICT ===\n")
         sys.stderr.flush()
+        
         return data_dict
 
 def get_raw_eval_results(results_path: str) -> list[EvalResult]:

src/populate.py

@@ -7,7 +7,8 @@ from src.leaderboard.read_evals import get_raw_eval_results
 def get_leaderboard_df(results_path: str, cols: list, benchmark_cols: list) -> pd.DataFrame:
     """Creates a dataframe from all the individual experiment results"""
     try:
-        sys.stderr.write("\n=== Starting leaderboard creation ===\n")
+        sys.stderr.write("\n=== GET_LEADERBOARD_DF DEBUG ===\n")
+        sys.stderr.write("Starting leaderboard creation...\n")
         sys.stderr.write(f"Looking for results in: {results_path}\n")
         sys.stderr.write(f"Expected columns: {cols}\n")
         sys.stderr.write(f"Benchmark columns: {benchmark_cols}\n")
@@ -17,81 +18,143 @@ def get_leaderboard_df(results_path: str, cols: list, benchmark_cols: list) -> p
         sys.stderr.write(f"\nFound {len(raw_data)} raw results\n")
         sys.stderr.flush()
 
+        if not raw_data:
+            sys.stderr.write("No raw data found, creating empty DataFrame\n")
+            sys.stderr.flush()
+            return create_empty_dataframe(cols, benchmark_cols)
+
         all_data_json = []
         for i, v in enumerate(raw_data):
             try:
+                sys.stderr.write(f"Processing result {i+1}/{len(raw_data)}: {v.full_model}\n")
+                sys.stderr.flush()
+                
                 data_dict = v.to_dict()
+                
+                # Validate the data_dict has required columns
+                missing_cols = [col for col in cols if col not in data_dict]
+                if missing_cols:
+                    sys.stderr.write(f"WARNING: Result for {v.full_model} missing columns: {missing_cols}\n")
+                    # Add missing columns with default values
+                    for col in missing_cols:
+                        if col in benchmark_cols or col == AutoEvalColumn.average.name:
+                            data_dict[col] = 0.0
+                        elif col == AutoEvalColumn.model_type_symbol.name:
+                            data_dict[col] = "?"
+                        else:
+                            data_dict[col] = ""
+                    sys.stderr.flush()
+                
                 all_data_json.append(data_dict)
                 sys.stderr.write(f"Successfully processed result {i+1}/{len(raw_data)}: {v.full_model}\n")
                 sys.stderr.flush()
+                
             except Exception as e:
                 sys.stderr.write(f"Error processing result {i+1}/{len(raw_data)} ({v.full_model}): {e}\n")
+                import traceback
+                sys.stderr.write(f"Traceback: {traceback.format_exc()}\n")
                 sys.stderr.flush()
                 continue
         
         sys.stderr.write(f"\nConverted to {len(all_data_json)} JSON records\n")
         sys.stderr.flush()
         
+        if not all_data_json:
+            sys.stderr.write("No valid JSON records, creating empty DataFrame\n")
+            sys.stderr.flush()
+            return create_empty_dataframe(cols, benchmark_cols)
+        
         if all_data_json:
             sys.stderr.write("Sample record keys: " + str(list(all_data_json[0].keys())) + "\n")
             sys.stderr.flush()
 
-        if not all_data_json:
-            sys.stderr.write("\nNo data found, creating empty DataFrame\n")
+        try:
+            df = pd.DataFrame.from_records(all_data_json)
+            sys.stderr.write("\nCreated DataFrame with columns: " + str(df.columns.tolist()) + "\n")
+            sys.stderr.write("DataFrame shape: " + str(df.shape) + "\n")
             sys.stderr.flush()
-            empty_df = pd.DataFrame(columns=cols)
-            # Ensure correct column types
-            empty_df[AutoEvalColumn.average.name] = pd.Series(dtype=float)
-            for col in benchmark_cols:
-                empty_df[col] = pd.Series(dtype=float)
-            return empty_df
-
-        df = pd.DataFrame.from_records(all_data_json)
-        sys.stderr.write("\nCreated DataFrame with columns: " + str(df.columns.tolist()) + "\n")
-        sys.stderr.write("DataFrame shape: " + str(df.shape) + "\n")
-        sys.stderr.flush()
+        except Exception as e:
+            sys.stderr.write(f"Error creating DataFrame from records: {e}\n")
+            sys.stderr.flush()
+            return create_empty_dataframe(cols, benchmark_cols)
 
         try:
-            df = df.sort_values(by=[AutoEvalColumn.average.name], ascending=False)
-            sys.stderr.write("\nSorted DataFrame by average\n")
+            if AutoEvalColumn.average.name in df.columns:
+                df = df.sort_values(by=[AutoEvalColumn.average.name], ascending=False)
+                sys.stderr.write("\nSorted DataFrame by average\n")
+            else:
+                sys.stderr.write(f"\nWARNING: Cannot sort by {AutoEvalColumn.average.name} - column not found\n")
             sys.stderr.flush()
-        except KeyError as e:
+        except Exception as e:
             sys.stderr.write(f"\nError sorting DataFrame: {e}\n")
             sys.stderr.write("Available columns: " + str(df.columns.tolist()) + "\n")
             sys.stderr.flush()
 
         try:
+            # Ensure all required columns exist before selecting
+            for col in cols:
+                if col not in df.columns:
+                    sys.stderr.write(f"Adding missing column during selection: {col}\n")
+                    if col in benchmark_cols or col == AutoEvalColumn.average.name:
+                        df[col] = 0.0
+                    else:
+                        df[col] = ""
+                    sys.stderr.flush()
+            
             df = df[cols].round(decimals=2)
             sys.stderr.write("\nSelected and rounded columns\n")
             sys.stderr.flush()
-        except KeyError as e:
+        except Exception as e:
             sys.stderr.write(f"\nError selecting columns: {e}\n")
             sys.stderr.write("Requested columns: " + str(cols) + "\n")
             sys.stderr.write("Available columns: " + str(df.columns.tolist()) + "\n")
             sys.stderr.flush()
-            # Create empty DataFrame with correct structure
-            empty_df = pd.DataFrame(columns=cols)
-            empty_df[AutoEvalColumn.average.name] = pd.Series(dtype=float)
-            for col in benchmark_cols:
-                empty_df[col] = pd.Series(dtype=float)
-            return empty_df
+            return create_empty_dataframe(cols, benchmark_cols)
 
-        # filter out if perplexity hasn't been evaluated
-        df = df[has_no_nan_values(df, benchmark_cols)]
-        sys.stderr.write("\nFinal DataFrame shape after filtering: " + str(df.shape) + "\n")
-        sys.stderr.write("Final columns: " + str(df.columns.tolist()) + "\n")
-        sys.stderr.flush()
+        try:
+            # filter out if perplexity hasn't been evaluated
+            df = df[has_no_nan_values(df, benchmark_cols)]
+            sys.stderr.write("\nFinal DataFrame shape after filtering: " + str(df.shape) + "\n")
+            sys.stderr.write("Final columns: " + str(df.columns.tolist()) + "\n")
+            sys.stderr.flush()
+        except Exception as e:
+            sys.stderr.write(f"Error filtering DataFrame: {e}\n")
+            sys.stderr.flush()
+            # Don't return empty, return the unfiltered DataFrame
         
+        # Final validation
+        if df is None or df.empty:
+            sys.stderr.write("Final DataFrame is None or empty, returning fallback\n")
+            sys.stderr.flush()
+            return create_empty_dataframe(cols, benchmark_cols)
+        
+        sys.stderr.write(f"=== FINAL RESULT: DataFrame with {len(df)} rows and {len(df.columns)} columns ===\n")
+        sys.stderr.flush()
         return df
         
     except Exception as e:
-        sys.stderr.write(f"\nCritical error in get_leaderboard_df: {e}\n")
+        sys.stderr.write(f"\nCRITICAL ERROR in get_leaderboard_df: {e}\n")
         import traceback
         sys.stderr.write(f"Traceback: {traceback.format_exc()}\n")
         sys.stderr.flush()
-        # Return empty DataFrame as fallback
-        empty_df = pd.DataFrame(columns=cols)
-        empty_df[AutoEvalColumn.average.name] = pd.Series(dtype=float)
-        for col in benchmark_cols:
+        # Always return a valid DataFrame, never None
+        return create_empty_dataframe(cols, benchmark_cols)
+
+def create_empty_dataframe(cols: list, benchmark_cols: list) -> pd.DataFrame:
+    """Create a valid empty DataFrame with all required columns"""
+    import sys
+    
+    sys.stderr.write("Creating empty fallback DataFrame...\n")
+    sys.stderr.flush()
+    
+    empty_df = pd.DataFrame(columns=cols)
+    # Ensure correct column types
+    for col in cols:
+        if col in benchmark_cols or col == AutoEvalColumn.average.name:
             empty_df[col] = pd.Series(dtype=float)
-        return empty_df
+        else:
+            empty_df[col] = pd.Series(dtype=str)
+    
+    sys.stderr.write(f"Empty DataFrame created with columns: {empty_df.columns.tolist()}\n")
+    sys.stderr.flush()
+    return empty_df