add overview + more info

app.py

@@ -171,7 +171,11 @@ class LeaderboardData:
         """
         local_df = eval_results[split]
         local_df = local_df.map(
-            lambda row: {"Model": model_hyperlink(row["url"], row["Model"])}
+            lambda row: {
+                "Model": model_hyperlink(
+                    row["model_url"], row["paper_url"], row["Model"]
+                )
+            }
         )
         filtered_columns = (
             PRE_COLUMN_NAMES
@@ -195,8 +199,9 @@ class LeaderboardData:
         df[f"{subsplit}_forces_mae"] = df[f"{subsplit}_forces_mae"]
         df = df.rename(
             columns={
-                f"{subsplit}_energy_mae": "Energy MAE [meV]",
-                f"{subsplit}_forces_mae": "Forces MAE [meV/Å]",
+                f"{subsplit}_energy_mae": "Energy MAE\n[meV]",
+                f"{subsplit}_forces_mae": "Forces MAE\n[meV/Å]",
+                "Energy Conserving": "Energy\nConserving",
             }
         )
         return df
@@ -207,7 +212,11 @@ class LeaderboardData:
         """
         local_df = eval_results[split]
         local_df = local_df.map(
-            lambda row: {"Model": model_hyperlink(row["url"], row["Model"])}
+            lambda row: {
+                "Model": model_hyperlink(
+                    row["model_url"], row["paper_url"], row["Model"]
+                )
+            }
         )
         eval_columns = LEADERBOARD_COLUMNS[split]
         filtered_columns = PRE_COLUMN_NAMES + eval_columns + POST_COLUMN_NAMES
@@ -242,18 +251,19 @@ LEADERBOARD_COLUMNS = {
 }
 
 COLUMN_MAPPING = {
-    "interaction_energy_mae": "Ixn Energy MAE [meV]",
-    "interaction_forces_mae": "Ixn Forces MAE [meV/Å]",
-    "strain_energy_mae": "Strain Energy MAE [meV]",
-    "deltaE_mae": "\u0394Energy MAE [meV]",
-    "deltaF_mae": "\u0394Forces MAE [meV/Å]",
+    "interaction_energy_mae": "Ixn Energy\nMAE [meV]",
+    "interaction_forces_mae": "Ixn Forces\nMAE [meV/Å]",
+    "strain_energy_mae": "Strain Energy\nMAE [meV]",
+    "deltaE_mae": "\u0394Energy MAE\n[meV]",
+    "deltaF_mae": "\u0394Forces MAE\n[meV/Å]",
     "ensemble_rmsd": "RMSD [Å]",
     "global_min_rmsd": "RMSD [Å]",
     "rmsd": "RMSD [Å]",
-    "lr_ddE_mae": "\u0394Energy (LR) MAE [meV]",
-    "lr_ddF_mae": "\u0394Forces (LR) MAE [meV/Å]",
-    "sr_ddE_mae": "\u0394Energy (SR) MAE [meV]",
-    "sr_ddF_mae": "\u0394Forces (SR) MAE [meV/Å]",
+    "lr_ddE_mae": "\u0394Energy (LR)\n MAE [meV]",
+    "lr_ddF_mae": "\u0394Forces (LR)\n MAE [meV/Å]",
+    "sr_ddE_mae": "\u0394Energy (SR)\n MAE [meV]",
+    "sr_ddF_mae": "\u0394Forces (SR)\n MAE [meV/Å]",
+    "Energy Conserving": "Energy\nConserving",
 }
 
 
@@ -262,7 +272,9 @@ def add_new_eval(
     eval_type: str,
     organization: str,
     model: str,
-    url: str,
+    model_url: str,
+    paper_url: str,
+    energy_conserving: bool,
     mail: str,
     training_set: str,
     additional_info: str,
@@ -331,8 +343,10 @@ def add_new_eval(
             "Organization": organization,
             "Submission date": submission_time,
             "Training Set": training_set,
+            "Energy Conserving": "✅" if energy_conserving else "",
+            "model_url": model_url,
+            "paper_url": paper_url,
             "Notes": additional_info,
-            "url": url,
         }
         eval_entry.update(metrics)
 
@@ -388,7 +402,10 @@ def add_new_eval(
 
 
 def create_dataframe_tab(
-    tab_name: str, df: pd.DataFrame, datatype: List[str] = None
+    tab_name: str,
+    df: pd.DataFrame,
+    datatype: List[str] = None,
+    widths: List[str] = None,
 ) -> gr.Tab:
     """
     Create a tab with a dataframe.
@@ -396,12 +413,19 @@ def create_dataframe_tab(
     if datatype is None:
         datatype = TYPES
 
+    if widths is None:
+        num_cols = len(df.columns)
+        fixed_cols = len(PRE_COLUMN_NAMES) + len(POST_COLUMN_NAMES)
+        # Model | Organization |Energy Conserving | Training Set | Metrics | date
+        widths = ["10%", "5%", "5%", "5%"] + ["5%"] * (num_cols - fixed_cols) + ["10%"]
+
     with gr.Tab(tab_name) as tab:
         gr.Dataframe(
             value=df,
             datatype=datatype,
             interactive=False,
-            column_widths=["20%"],
+            show_search="filter",
+            column_widths=widths,
         )
     return tab
 
@@ -425,15 +449,121 @@ def create_s2ef_tabs(split: str, results_dfs: Dict[str, pd.DataFrame]) -> None:
 
 def create_evaluation_tabs(results_dfs: Dict[str, pd.DataFrame]) -> None:
     """
-    Create evaluation tabs for non-S2EF evaluations.
+    Create evaluation tabs for non-S2EF evaluations, including Overview tab.
     """
     eval_datatype = ["markdown", "markdown", "number", "str"]
 
+    # Create Overview tab first
+    overview_df = create_overview_dataframe(results_dfs)
+    create_dataframe_tab("Overview", overview_df, eval_datatype, widths=["10%"])
+
+    # Create individual evaluation tabs
     for eval_type in OTHER_EVAL_TYPES:
         display_name = "IE/EA" if eval_type == "IE_EA" else eval_type
         create_dataframe_tab(display_name, results_dfs[eval_type], eval_datatype)
 
 
+def create_overview_dataframe(results_dfs: Dict[str, pd.DataFrame]) -> pd.DataFrame:
+    """
+    Create an overview dataframe combining all models with only the first metric from each eval type.
+    """
+    # Initialize overview data with model info
+    overview_data = {}
+
+    # Get all unique model-dataset combinations across all dataframes
+    all_model_entries = set()
+    model_info = {}  # Store model metadata (org, dataset, etc.)
+
+    # Collect all models and their info from all evaluation types
+    for eval_type, df in results_dfs.items():
+        if eval_type.startswith("Validation_") or eval_type.startswith("Test_"):
+            continue
+
+        for _, row in df.iterrows():
+            model_name = row["Model"]
+            dataset = row["Training Set"]
+            # Create unique identifier combining model name and training set
+            model_entry = (model_name, dataset)
+            all_model_entries.add(model_entry)
+            # Store model metadata for this specific entry
+            model_info[model_entry] = {
+                "Model": model_name,
+                "Organization": row.get("Organization", ""),
+                "Energy Conserving": row.get("Energy\nConserving", ""),
+                "Training Set": dataset,
+            }
+
+    # Initialize overview data structure
+    overview_data = {
+        "Model": [],
+        "Organization": [],
+        "Energy Conserving": [],
+        "Training Set": [],
+    }
+
+    # Add columns for the primary metric from each evaluation type
+    metric_columns = {}
+
+    # Add primary metric from each OTHER evaluation type (skip S2EF)
+    for eval_type in OTHER_EVAL_TYPES:
+        if eval_type in results_dfs and eval_type in LEADERBOARD_COLUMNS:
+            primary_metric = LEADERBOARD_COLUMNS[eval_type][0]  # First metric
+            # Map to display name using COLUMN_MAPPING
+            metric_display_name = COLUMN_MAPPING.get(primary_metric, primary_metric)
+            # Include task name to avoid conflicts when multiple tasks have same metric
+            task_display_name = "IE/EA" if eval_type == "IE_EA" else eval_type
+            full_display_name = f"{task_display_name}\n{metric_display_name}"
+            overview_data[full_display_name] = []
+            metric_columns[full_display_name] = (eval_type, metric_display_name)
+
+    # Populate data for each model entry
+    for model_entry in sorted(
+        all_model_entries, key=lambda x: (x[0], x[1])
+    ):  # Sort by model name, then dataset
+        model_name, dataset = model_entry
+        entry_info = model_info[model_entry]
+
+        overview_data["Model"].append(entry_info["Model"])
+        overview_data["Organization"].append(entry_info["Organization"])
+        overview_data["Energy Conserving"].append(entry_info["Energy Conserving"])
+        overview_data["Training Set"].append(entry_info["Training Set"])
+
+        # Fill in metrics for each column
+        for display_col, (eval_type, source_col) in metric_columns.items():
+            if eval_type in results_dfs:
+                df = results_dfs[eval_type]
+                # Match both model name and training set
+                model_row = df[
+                    (df["Model"] == model_name) & (df["Training Set"] == dataset)
+                ]
+                if not model_row.empty and source_col in model_row.columns:
+                    value = model_row.iloc[0][source_col]
+                else:
+                    value = "-"
+            else:
+                value = "-"
+            overview_data[display_col].append(value)
+
+    overview_df = pd.DataFrame(overview_data)
+
+    # Sort by the average of all metric columns (ascending for MAE metrics)
+    metric_cols = [
+        col
+        for col in overview_df.columns
+        if col not in PRE_COLUMN_NAMES + POST_COLUMN_NAMES
+    ]
+    if metric_cols:
+        # Calculate average across all metric columns for each row
+        # Convert all metric columns to numeric, keeping "-" as NaN
+        numeric_metrics = overview_df[metric_cols].apply(pd.to_numeric, errors="coerce")
+        # Calculate mean across columns, ignoring NaN values
+        avg_scores = numeric_metrics.mean(axis=1)
+        # Sort by average score (ascending for MAE metrics)
+        overview_df = overview_df.loc[avg_scores.sort_values(na_position="last").index]
+
+    return overview_df
+
+
 def create_submission_interface() -> Tuple[gr.components.Component, ...]:
     """
     Create the submission interface components.
@@ -444,7 +574,11 @@ def create_submission_interface() -> Tuple[gr.components.Component, ...]:
         with gr.Row():
             with gr.Column():
                 model_name_textbox = gr.Textbox(label="Model name")
-                model_url = gr.Textbox(label="Model/Paper URL")
+                energy_conserving = gr.Checkbox(
+                    label="Is the model energy conserving? (i.e. F= -dE/dx)"
+                )
+                model_url = gr.Textbox(label="Model/Checkpoint URL")
+                paper_url = gr.Textbox(label="Paper URL")
                 dataset = gr.Dropdown(
                     choices=["OMol-All", "OMol-4M", "UMA-459M", "Other"],
                     label="Training set",
@@ -478,6 +612,8 @@ def create_submission_interface() -> Tuple[gr.components.Component, ...]:
         organization,
         model_name_textbox,
         model_url,
+        paper_url,
+        energy_conserving,
         mail,
         dataset,
         additional_info,
@@ -523,6 +659,8 @@ def create_interface() -> gr.Blocks:
             organization,
             model_name_textbox,
             model_url,
+            paper_url,
+            energy_conserving,
             mail,
             dataset,
             additional_info,
@@ -531,7 +669,7 @@ def create_interface() -> gr.Blocks:
 
         submit_button.click(
             lambda x: "Coming soon :)",
-            None,
+            [0],
             submission_result,
             # add_new_eval,
             # [
@@ -545,7 +683,6 @@ def create_interface() -> gr.Blocks:
             #     additional_info,
             # ],
             # submission_result,
-            
         )
 
     return demo

content.py

@@ -60,13 +60,15 @@ CITATION_BUTTON_TEXT = r"""
 """
 
 # Table configuration
-PRE_COLUMN_NAMES = ["Model", "Organization", "Training Set"]
+PRE_COLUMN_NAMES = ["Model", "Organization", "Energy Conserving", "Training Set"]
 POST_COLUMN_NAMES = ["Submission date"]
-TYPES = ["markdown", "markdown", "str", "number", "number", "str"]
+TYPES = ["markdown", "str", "bool", "str"]
 
 
-def model_hyperlink(link: str, model_name: str) -> str:
+def model_hyperlink(model_link: str, paper_link: str, model_name: str) -> str:
     """Create a hyperlink for model names in the leaderboard."""
-    if not link or link.strip() == "":
+    if not model_link or model_link.strip() == "":
         return model_name
-    return f'<a target="_blank" href="{link}" style="color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;">{model_name}</a>'
+    if not paper_link or paper_link.strip() == "" or paper_link == "-":
+        return f'<a target="_blank" href="{model_link}" style="color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;" title="Download model">{model_name}</a>'
+    return f'<a target="_blank" href="{model_link}" style="color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;" title="Download model">{model_name}</a> <a target="_blank" href="{paper_link}" style="color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;" title="Read paper">📕</a>'