add longform tab

app.py

@@ -1,9 +1,9 @@
 import gradio as gr
 import pandas as pd
 import json
-from constants import BANNER, INTRODUCTION_TEXT, CITATION_TEXT, METRICS_TAB_TEXT, DIR_OUTPUT_REQUESTS, LEADERBOARD_CSS, EU_LANGUAGES, MULTILINGUAL_TAB_TEXT
+from constants import BANNER, INTRODUCTION_TEXT, CITATION_TEXT, METRICS_TAB_TEXT, DIR_OUTPUT_REQUESTS, LEADERBOARD_CSS, EU_LANGUAGES, MULTILINGUAL_TAB_TEXT, LONGFORM_TAB_TEXT
 from init import is_model_on_hub, upload_file, load_all_info_from_dataset_hub
-from utils_display import AutoEvalColumn, MultilingualColumn, fields, make_clickable_model, styled_error, styled_message
+from utils_display import AutoEvalColumn, MultilingualColumn, LongformColumn, fields, make_clickable_model, styled_error, styled_message
 import numpy as np
 from datetime import datetime, timezone
 
@@ -27,7 +27,7 @@ column_names = {
     "Voxpopuli WER": "Voxpopuli",
 }
 
-eval_queue_repo, requested_models, csv_results, multilingual_csv_path = load_all_info_from_dataset_hub()
+eval_queue_repo, requested_models, csv_results, multilingual_csv_path, longform_csv_path = load_all_info_from_dataset_hub()
 
 if not csv_results.exists():
     raise Exception(f"CSV file {csv_results} does not exist locally")
@@ -57,6 +57,10 @@ TYPES = [c.type for c in fields(AutoEvalColumn)]
 # Multilingual columns (dynamic based on expansion state)
 MULTILINGUAL_COLS = [c.name for c in fields(MultilingualColumn)]
 
+# Longform columns
+LONGFORM_COLS = [c.name for c in fields(LongformColumn)]
+LONGFORM_TYPES = [c.type for c in fields(LongformColumn)]
+
 def create_multilingual_dataframe():
     """Create multilingual dataframe with CoVoST, MLS, and FLEURS benchmark data"""
     global benchmark_details, expanded_languages
@@ -225,6 +229,43 @@ def toggle_language_expansion(language_code):
 # Initialize multilingual dataframe
 multilingual_df = create_multilingual_dataframe()
 
+def create_longform_dataframe():
+    """Create longform dataframe from CSV data"""
+    if longform_csv_path is not None and longform_csv_path.exists():
+        longform_raw_df = pd.read_csv(longform_csv_path)
+        longform_data = []
+        
+        for _, row_data in longform_raw_df.iterrows():
+            model_name = row_data['model_id']
+            
+            # Get values from CSV, similar to other tabs
+            earnings21_wer = row_data.get('earnings21', -1)
+            earnings22_wer = row_data.get('earnings22', -1)
+            tedlium_wer = row_data.get('tedlium', -1)
+            rtfx_value = row_data.get('RTFx', 0)
+            
+            # Calculate average WER from available datasets
+            available_wers = [w for w in [earnings21_wer, tedlium_wer] if w != -1 and w > 0]
+            avg_wer = round(np.mean(available_wers), 2) if available_wers else 0.0
+            
+            row = {
+                "Model": make_clickable_model(model_name),
+                "Average WER ⬇️": avg_wer,
+                "RTFx ⬆️️": rtfx_value if rtfx_value > 0 else "NA",
+                "Earnings21": earnings21_wer if earnings21_wer != -1 else "NA",
+                "Earnings22": earnings22_wer if earnings22_wer != -1 else "NA",
+                "Tedlium": tedlium_wer if tedlium_wer != -1 else "NA",
+            }
+            longform_data.append(row)
+        
+        longform_df = pd.DataFrame(longform_data)
+    
+    longform_df = longform_df.sort_values(by='Average WER ⬇️')
+    return longform_df
+
+# Initialize longform dataframe
+longform_df = create_longform_dataframe()
+
 
 def request_model(model_text, chbcoco2017):
     
@@ -363,10 +404,21 @@ with gr.Blocks(css=LEADERBOARD_CSS) as demo:
                     outputs=[multilingual_table]
                 )
 
-        with gr.TabItem("📈 Metrics", elem_id="od-benchmark-tab-table", id=3):
+        with gr.TabItem("📝 Long-form", elem_id="longform-benchmark-tab-table", id=2):
+            gr.Markdown(LONGFORM_TAB_TEXT, elem_classes="markdown-text")
+            
+            longform_table = gr.components.Dataframe(
+                value=longform_df,
+                datatype=LONGFORM_TYPES,
+                elem_id="longform-table",
+                interactive=False,
+                visible=True,
+            )
+
+        with gr.TabItem("📈 Metrics", elem_id="od-benchmark-tab-table", id=4):
             gr.Markdown(METRICS_TAB_TEXT, elem_classes="markdown-text")
 
-        with gr.TabItem("✉️✨ Request a model here!", elem_id="od-benchmark-tab-table", id=4):
+        with gr.TabItem("✉️✨ Request a model here!", elem_id="od-benchmark-tab-table", id=5):
             with gr.Column():
                 gr.Markdown("# ✉️✨ Request results for a new model here!", elem_classes="markdown-text")
             with gr.Column():
@@ -381,7 +433,7 @@ with gr.Blocks(css=LEADERBOARD_CSS) as demo:
                                       [model_name_textbox, chb_coco2017], 
                                       mdw_submission_result)
         # add an about section
-        with gr.TabItem("🤗 About", elem_id="od-benchmark-tab-table", id=5):
+        with gr.TabItem("🤗 About", elem_id="od-benchmark-tab-table", id=6):
             gr.Markdown("## About", elem_classes="markdown-text")
 
     gr.Markdown(f"Last updated on **{LAST_UPDATED}**", elem_classes="markdown-text")