Model type filter

app.py

@@ -32,6 +32,7 @@ HF_TOKEN = os.getenv("HF_TOKEN")
 RESULTS_LOCK = threading.Lock()
 DEFAULT_METRIC = "NDCG@10"
 METRICS = ["NDCG@10", "MAP@10", "Recall@10", "P@10", "MRR@10"]
+RERANKER_FILTER_CHOICES = ["All models", "With reranker", "Single models"]
 STATIC_COLUMNS = ["Rank", "Model"]
 META_COLUMNS = ["Model ID", "Organization", "Type", "Verified", "Date", "Source URL", ]
 TRAILING_COLUMNS = META_COLUMNS
@@ -716,8 +717,8 @@ def sort_type_for_column(index: int, column: str, metric: str) -> str:
     return "text"
 
 
-def leaderboard_table_html(metric: str, task_filter: str, verified_only: bool, model_filter: str) -> str:
-    frame = leaderboard_frame(metric, task_filter, verified_only, model_filter)
+def leaderboard_table_html(metric: str, task_filter: str, verified_only: bool, model_filter: str, reranker_filter: str = "All models") -> str:
+    frame = leaderboard_frame(metric, task_filter, verified_only, model_filter, reranker_filter)
     if frame.empty:
         return '<div class="rusbeir-empty">No results match the selected filters.</div>'
 
@@ -920,21 +921,16 @@ def append_submission_records(records: list[dict[str, Any]]) -> tuple[int, int,
     raise RuntimeError("Could not update leaderboard results after multiple attempts.")
 
 
-def add_submission_record(uploaded_file: Any, metric: str, task_filter: str, verified_only: bool, model_filter: str) -> tuple[str, str, str]:
+def add_submission_record(uploaded_file: Any, metric: str, task_filter: str, verified_only: bool, model_filter: str,  reranker_filter: str = "All models") -> tuple[str, str, str]:
     try:
         file_text = uploaded_file_text(uploaded_file)
     except OSError as exc:
-        return (
-            f"Submission was not added: could not read uploaded file: {exc}",
-            summary_html(),
-            leaderboard_table_html(metric, task_filter, verified_only, model_filter),
-        )
+        return (f"Submission was not added: could not read uploaded file: {exc}", summary_html(),
+            leaderboard_table_html(metric, task_filter, verified_only, model_filter, reranker_filter))
+    
     if not file_text.strip():
-        return (
-            "Upload a non-empty results.jsonl file first.",
-            summary_html(),
-            leaderboard_table_html(metric, task_filter, verified_only, model_filter),
-        )
+        return ("Upload a non-empty results.jsonl file first.", summary_html(),
+            leaderboard_table_html(metric, task_filter, verified_only, model_filter, reranker_filter))
 
     try:
         records = parse_submission_records(file_text)
@@ -943,7 +939,7 @@ def add_submission_record(uploaded_file: Any, metric: str, task_filter: str, ver
         return (
             f"Submission was not added: {exc}",
             summary_html(),
-            leaderboard_table_html(metric, task_filter, verified_only, model_filter),
+            leaderboard_table_html(metric, task_filter, verified_only, model_filter, reranker_filter),
         )
 
     if added == 0 and skipped > 0:
@@ -955,11 +951,8 @@ def add_submission_record(uploaded_file: Any, metric: str, task_filter: str, ver
         if commit_url:
             status += f" [Hub commit]({commit_url})"
 
-    return (
-        status,
-        summary_html(),
-        leaderboard_table_html(metric, task_filter, verified_only, model_filter),
-    )
+    return (status, summary_html(),
+        leaderboard_table_html(metric, task_filter, verified_only, model_filter, reranker_filter))
 
 
 def logo_data_uri() -> str:
@@ -982,7 +975,7 @@ def format_score(value: float | None) -> str:
 def summary_html() -> str:
     datasets = [dataset for dataset in load_datasets() if dataset.get("official", True)]
     records = load_results()
-    frame = leaderboard_frame(DEFAULT_METRIC, "All", False, "")
+    frame = leaderboard_frame(DEFAULT_METRIC, "All", False, "", "All models")
     best_model = "No results yet"
     best_score = "n/a"
     if not frame.empty and DEFAULT_METRIC in frame:
@@ -1040,7 +1033,11 @@ def summary_html() -> str:
     """
 
 
-def leaderboard_frame(metric: str, task_filter: str, verified_only: bool, model_filter: str) -> pd.DataFrame:
+def leaderboard_frame(metric: str, task_filter: str,
+    verified_only: bool,
+    model_filter: str,
+    reranker_filter: str = "All models",
+) -> pd.DataFrame:
     datasets = load_datasets()
     if task_filter != "All":
         datasets = [dataset for dataset in datasets if dataset["task"] == task_filter]
@@ -1050,6 +1047,11 @@ def leaderboard_frame(metric: str, task_filter: str, verified_only: bool, model_
     for record in load_results():
         if verified_only and not record.get("verified", False):
             continue
+        has_reranker = str(record.get("type", "")).strip().casefold() == "reranker"
+        if reranker_filter == "With reranker" and not has_reranker:
+            continue
+        if reranker_filter == "Single models" and has_reranker:
+            continue
         model_text = f"{record.get('model_id', '')} {record.get('model_name', '')}".lower()
         if model_filter and model_filter.lower() not in model_text:
             continue
@@ -1115,7 +1117,7 @@ with gr.Blocks(title="rusBEIR Leaderboard") as demo:
                     """
                     <h2 class="rusbeir-section-title">Model Rankings</h2>
                     <p class="rusbeir-section-note">
-                      Filter by task family, model name, or verification status. Scores are stored as fractions;
+                      Filter by task family, model name, reranker usage, or verification status. Scores are stored as fractions;
                       leaderboard rankings use the selected average metric.
                     </p>
                     """
@@ -1124,19 +1126,20 @@ with gr.Blocks(title="rusBEIR Leaderboard") as demo:
                     metric = gr.Dropdown(METRICS, value=DEFAULT_METRIC, label="Metric", min_width=150)
                     task_filter = gr.Dropdown(task_choices(), value="All", label="Task", min_width=180)
                     model_filter = gr.Textbox(label="Model", placeholder="intfloat, BGE, FRIDA...", min_width=220)
+                    reranker_filter = gr.Dropdown(RERANKER_FILTER_CHOICES, value="All models", label="Reranker", min_width=170)
                     verified_only = gr.Checkbox(value=False, label="Verified", min_width=120, elem_classes=["rusbeir-verified-filter"])
 
                 gr.HTML('<div class="rusbeir-section-note">Results</div>')
-                table = gr.HTML(value=leaderboard_table_html(DEFAULT_METRIC, "All", False, ""))
+                table = gr.HTML(value=leaderboard_table_html(DEFAULT_METRIC, "All", False, "", "All models"))
 
-                for control in [metric, task_filter, verified_only, model_filter]:
+                for control in [metric, task_filter, verified_only, model_filter, reranker_filter]:
                     control.change(leaderboard_table_html,
-                        inputs=[metric, task_filter, verified_only, model_filter],
+                        inputs=[metric, task_filter, verified_only, model_filter, reranker_filter],
                         outputs=table)
 
                 reload_button = gr.Button("Reload results", variant="secondary")
                 reload_button.click(leaderboard_table_html,
-                    inputs=[metric, task_filter, verified_only, model_filter],
+                    inputs=[metric, task_filter, verified_only, model_filter, reranker_filter],
                     outputs=table)
 
         with gr.Tab("Datasets"):
@@ -1179,7 +1182,7 @@ with gr.Blocks(title="rusBEIR Leaderboard") as demo:
                 submit_status = gr.Markdown()
                 add_button = gr.Button("Add to leaderboard", variant="primary")
                 add_button.click(add_submission_record,
-                    inputs=[results_file, metric, task_filter, verified_only, model_filter],
+                    inputs=[results_file, metric, task_filter, verified_only, model_filter, reranker_filter],
                     outputs=[submit_status, summary, table])
 
         with gr.Tab("About"):