import gradio as gr import pandas as pd from apscheduler.schedulers.background import BackgroundScheduler from src.about import ( CITATION_BUTTON_LABEL, CITATION_BUTTON_TEXT, INTRODUCTION_TEXT, LLM_BENCHMARKS_TEXT, TITLE, ) from src.display.css_html_js import custom_css from src.display.formatting import get_leaderboard_table_html from src.envs import API, REPO_ID, TOKEN # ============================================================================= # Helper Functions # ============================================================================= def restart_space(): """Restart the Hugging Face space""" API.restart_space(repo_id=REPO_ID) def load_medical_data(): """Load and preprocess medical LLM evaluation data""" try: df = pd.read_csv("data/llm_medical_model_M80_comparison.csv") df_ability = pd.read_csv("data/model_abilities_by_topic.csv") topics = [l.strip() for l in open("data/llm_medical_multi_run_topics.txt", encoding="utf-8") if l.strip()] # Medical specialty column mappings column_renames = { 'acc_Behavioral Health & Nervous Systems/Special Senses': 'Neuro', 'acc_Biostatistics & Epidemiology/Population Health': 'Biostats', 'acc_Blood & Lymphoreticular/Immune Systems': 'Hemo', 'acc_Cardiovascular System': 'Cardio', 'acc_Gastrointestinal System': 'GI', 'acc_Human Development': 'DevMed', 'acc_Multisystem Processes & Disorders': 'MultiSys', 'acc_Musculoskeletal, Skin & Subcutaneous Tissue': 'MSK', 'acc_Reproductive & Endocrine Systems': 'Repro', 'acc_Respiratory & Renal/Urinary Systems': 'Resp', 'acc_Social Sciences: Communication and Interpersonal Skills': 'Comm' } abbr = ["Comm", "Behav/Neuro", "Blood/Immune", "MSK/Skin", "Multi", "GI", "Dev", "Resp/Renal", "Cardio", "Repro/Endo", "Bio/Epi"] df = df.rename(columns=column_renames) column_renames_ability = { f"irt_ability_{i+1}": topics[i] for i in range(len(topics)) } df_ability = df_ability.rename(columns=column_renames_ability) new_cols = ["model_id", "Mean"]+ [v for v in column_renames_ability.values()] df_ability = df_ability[new_cols] # Sort by accuracy and add ranking # df = df.sort_values('accuracy', ascending=False) # df.insert(0, 'rank', range(1, len(df) + 1)) df_ability = df_ability.sort_values('Mean', ascending=False) df_ability.insert(0, 'rank', range(1, len(df_ability) + 1)) return df_ability except Exception as e: print(f"Error loading medical data: {e}") return pd.DataFrame() def init_leaderboard(dataframe): """Initialize leaderboard with custom HTML table""" if dataframe is None or dataframe.empty: raise ValueError("Leaderboard DataFrame is empty or None.") return get_leaderboard_table_html(dataframe) # ============================================================================= # Load Data # ============================================================================= LEADERBOARD_DF = load_medical_data() # ============================================================================= # Gradio Interface # ============================================================================= demo = gr.Blocks(css=custom_css) with demo: # Header gr.HTML(TITLE) gr.Markdown(INTRODUCTION_TEXT, elem_classes="markdown-text") with gr.Tabs(elem_classes="tab-buttons") as tabs: # Main Leaderboard Tab with gr.TabItem("🏥 LLM Leaderboard for Medical Abilities", elem_id="medical-llm-tab", id=0): leaderboard_html = init_leaderboard(LEADERBOARD_DF) gr.HTML(leaderboard_html) # Radar Charts Tab with gr.TabItem("🎯 Ability Radar Charts", elem_id="radar-tab", id=1): gr.Markdown("", elem_classes="markdown-text-radar") # About Tab with gr.TabItem("📝 About", elem_id="llm-benchmark-tab-table", id=2): gr.Markdown(LLM_BENCHMARKS_TEXT, elem_classes="markdown-text") # Citation Footer with gr.Row(): with gr.Accordion("📙 Citation", open=False): citation_button = gr.Textbox( value=CITATION_BUTTON_TEXT, label=CITATION_BUTTON_LABEL, lines=20, elem_id="citation-button", show_copy_button=True, ) # ============================================================================= # Application Launch # ============================================================================= # Schedule automatic space restart every 30 minutes scheduler = BackgroundScheduler() scheduler.add_job(restart_space, "interval", seconds=1800) scheduler.start() # Launch the application demo.queue(default_concurrency_limit=40).launch()