import gradio as gr from apscheduler.schedulers.background import BackgroundScheduler from huggingface_hub import snapshot_download from src.about import ( INTRODUCTION_TEXT, BENCHMARKS_TEXT, TITLE, EVALUATION_QUEUE_TEXT ) from src.benchmarks import DOMAIN_COLS_QA, LANG_COLS_QA, DOMAIN_COLS_LONG_DOC, LANG_COLS_LONG_DOC, METRIC_LIST, \ DEFAULT_METRIC from src.display.css_html_js import custom_css from src.display.utils import COL_NAME_IS_ANONYMOUS, COL_NAME_REVISION, COL_NAME_TIMESTAMP from src.envs import API, EVAL_RESULTS_PATH, REPO_ID, RESULTS_REPO, TOKEN from src.read_evals import get_raw_eval_results, get_leaderboard_df from src.utils import update_table, update_metric, update_table_long_doc, upload_file, get_default_cols, submit_results, clear_reranking_selections def restart_space(): API.restart_space(repo_id=REPO_ID) try: snapshot_download( repo_id=RESULTS_REPO, local_dir=EVAL_RESULTS_PATH, repo_type="dataset", tqdm_class=None, etag_timeout=30, token=TOKEN ) except Exception as e: print(f'failed to download') restart_space() raw_data = get_raw_eval_results(f"{EVAL_RESULTS_PATH}/AIR-Bench_24.04") original_df_qa = get_leaderboard_df( raw_data, task='qa', metric=DEFAULT_METRIC) original_df_long_doc = get_leaderboard_df( raw_data, task='long-doc', metric=DEFAULT_METRIC) print(f'raw data: {len(raw_data)}') print(f'QA data loaded: {original_df_qa.shape}') print(f'Long-Doc data loaded: {len(original_df_long_doc)}') leaderboard_df_qa = original_df_qa.copy() # leaderboard_df_qa = leaderboard_df_qa[has_no_nan_values(df, _benchmark_cols)] shown_columns_qa, types_qa = get_default_cols( 'qa', leaderboard_df_qa.columns, add_fix_cols=True) leaderboard_df_qa = leaderboard_df_qa[~leaderboard_df_qa[COL_NAME_IS_ANONYMOUS]][shown_columns_qa] leaderboard_df_qa.drop([COL_NAME_REVISION, COL_NAME_TIMESTAMP], axis=1, inplace=True) leaderboard_df_long_doc = original_df_long_doc.copy() shown_columns_long_doc, types_long_doc = get_default_cols( 'long-doc', leaderboard_df_long_doc.columns, add_fix_cols=True) leaderboard_df_long_doc = leaderboard_df_long_doc[~leaderboard_df_long_doc[COL_NAME_IS_ANONYMOUS]][shown_columns_long_doc] leaderboard_df_long_doc.drop([COL_NAME_REVISION, COL_NAME_TIMESTAMP], axis=1, inplace=True) def update_metric_qa( metric: str, domains: list, langs: list, reranking_model: list, query: str, show_anonymous: bool, show_revision_and_timestamp, ): return update_metric(raw_data, 'qa', metric, domains, langs, reranking_model, query, show_anonymous, show_revision_and_timestamp) def update_metric_long_doc( metric: str, domains: list, langs: list, reranking_model: list, query: str, show_anonymous: bool, show_revision_and_timestamp, ): return update_metric(raw_data, "long-doc", metric, domains, langs, reranking_model, query, show_anonymous, show_revision_and_timestamp) def update_table_without_ranking( hidden_df, domains, langs, reranking_query, query, show_anonymous, show_revision_and_timestamp, ): return update_table(hidden_df, domains, langs, reranking_query, query, show_anonymous, reset_ranking=False, show_revision_and_timestamp=show_revision_and_timestamp) def update_table_without_ranking_long_doc( hidden_df, domains, langs, reranking_query, query, show_anonymous, show_revision_and_timestamp, ): return update_table_long_doc(hidden_df, domains, langs, reranking_query, query, show_anonymous, reset_ranking=False, show_revision_and_timestamp=show_revision_and_timestamp) demo = gr.Blocks(css=custom_css) with demo: gr.HTML(TITLE) gr.Markdown(INTRODUCTION_TEXT, elem_classes="markdown-text") with gr.Tabs(elem_classes="tab-buttons") as tabs: with gr.TabItem("QA", elem_id="qa-benchmark-tab-table", id=0): with gr.Row(): with gr.Column(): # search retrieval models with gr.Row(): selected_version = gr.Dropdown( choices=["AIR-Bench_24.04",], value="AIR-Bench_24.04", label="Select the version of AIR-Bench", interactive = True ) with gr.Row(): search_bar = gr.Textbox( placeholder=" 🔍 Search for retrieval models (separate multiple queries with `;`) and press ENTER...", show_label=False, elem_id="search-bar", info="Search the retrieval models" ) # select reranking model reranking_models = list(frozenset([eval_result.reranking_model for eval_result in raw_data])) with gr.Row(): selected_rerankings = gr.CheckboxGroup( choices=reranking_models, value=reranking_models, label="Select the reranking models", elem_id="reranking-select", interactive=True ) with gr.Row(): select_noreranker_only_btn = gr.ClearButton( selected_rerankings, value="Only show results without ranking models", ) with gr.Column(min_width=320): # select the metric selected_metric = gr.Dropdown( choices=METRIC_LIST, value=DEFAULT_METRIC, label="Select the metric", interactive=True, elem_id="metric-select", ) # select domain with gr.Row(): selected_domains = gr.CheckboxGroup( choices=DOMAIN_COLS_QA, value=DOMAIN_COLS_QA, label="Select the domains", elem_id="domain-column-select", interactive=True, ) # select language with gr.Row(): selected_langs = gr.Dropdown( choices=LANG_COLS_QA, value=LANG_COLS_QA, label="Select the languages", elem_id="language-column-select", multiselect=True, interactive=True ) with gr.Row(): show_anonymous = gr.Checkbox( label="Show anonymous submissions", value=False, info="The anonymous submissions might have invalid model information." ) with gr.Row(): show_revision_and_timestamp = gr.Checkbox( label="Show submission details", value=False, info="Show the revision and timestamp information of submissions" ) leaderboard_table = gr.components.Dataframe( value=leaderboard_df_qa, datatype=types_qa, elem_id="leaderboard-table", interactive=False, visible=True, ) # Dummy leaderboard for handling the case when the user uses backspace key hidden_leaderboard_table_for_search = gr.components.Dataframe( value=original_df_qa, datatype=types_qa, visible=False, ) # Set search_bar listener search_bar.submit( update_table_without_ranking, [ hidden_leaderboard_table_for_search, selected_domains, selected_langs, selected_rerankings, search_bar, show_anonymous, ], leaderboard_table, ) for selector in [show_revision_and_timestamp, selected_rerankings]: selector.change( update_table_without_ranking, [ hidden_leaderboard_table_for_search, selected_domains, selected_langs, selected_rerankings, search_bar, show_anonymous, show_revision_and_timestamp ], leaderboard_table, queue=True ) # Set column-wise listener for selector in [ selected_domains, selected_langs, show_anonymous ]: selector.change( update_table, [ hidden_leaderboard_table_for_search, selected_domains, selected_langs, selected_rerankings, search_bar, show_anonymous, ], leaderboard_table, queue=True, ) # set metric listener selected_metric.change( update_metric_qa, [ selected_metric, selected_domains, selected_langs, selected_rerankings, search_bar, show_anonymous, ], leaderboard_table, queue=True ) select_noreranker_only_btn.click( clear_reranking_selections, outputs=selected_rerankings ) with gr.TabItem("Long Doc", elem_id="long-doc-benchmark-tab-table", id=1): with gr.Row(): with gr.Column(): with gr.Row(): selected_version = gr.Dropdown( choices=["AIR-Bench_24.04",], value="AIR-Bench_24.04", label="Select the version of AIR-Bench", interactive=True ) with gr.Row(): search_bar = gr.Textbox( info="Search the retrieval models", placeholder=" 🔍 Search for retrieval models (separate multiple queries with `;`) and press ENTER...", show_label=False, elem_id="search-bar-long-doc", ) # select reranking model reranking_models = list(frozenset([eval_result.reranking_model for eval_result in raw_data])) with gr.Row(): selected_rerankings = gr.CheckboxGroup( choices=reranking_models, value=reranking_models, label="Select the reranking models", elem_id="reranking-select-long-doc", interactive=True ) with gr.Row(): select_noreranker_only_btn = gr.ClearButton( selected_rerankings, value="Only show results without ranking models", ) with gr.Column(min_width=320): # select the metric with gr.Row(): selected_metric = gr.Dropdown( choices=METRIC_LIST, value=DEFAULT_METRIC, label="Select the metric", interactive=True, elem_id="metric-select-long-doc", ) # select domain with gr.Row(): selected_domains = gr.CheckboxGroup( choices=DOMAIN_COLS_LONG_DOC, value=DOMAIN_COLS_LONG_DOC, label="Select the domains", elem_id="domain-column-select-long-doc", interactive=True, ) # select language with gr.Row(): selected_langs = gr.Dropdown( choices=LANG_COLS_LONG_DOC, value=LANG_COLS_LONG_DOC, label="Select the languages", elem_id="language-column-select-long-doc", multiselect=True, interactive=True ) with gr.Row(): show_anonymous = gr.Checkbox( label="Show anonymous submissions", value=False, info="The anonymous submissions might have invalid model information." ) with gr.Row(): show_revision_and_timestamp = gr.Checkbox( label="Show submission details", value=False, info="Show the revision and timestamp information of submissions" ) leaderboard_table_long_doc = gr.components.Dataframe( value=leaderboard_df_long_doc, datatype=types_long_doc, elem_id="leaderboard-table-long-doc", interactive=False, visible=True, ) # Dummy leaderboard for handling the case when the user uses backspace key hidden_leaderboard_table_for_search = gr.components.Dataframe( value=original_df_long_doc, datatype=types_long_doc, visible=False, ) # Set search_bar listener search_bar.submit( update_table_without_ranking_long_doc, [ hidden_leaderboard_table_for_search, selected_domains, selected_langs, selected_rerankings, search_bar, show_anonymous, show_revision_and_timestamp ], leaderboard_table_long_doc, ) for selector in [show_revision_and_timestamp, selected_rerankings]: selector.change( update_table_without_ranking_long_doc, [ hidden_leaderboard_table_for_search, selected_domains, selected_langs, selected_rerankings, search_bar, show_anonymous, show_revision_and_timestamp ], leaderboard_table_long_doc, queue=True, ) # Set column-wise listener for selector in [ selected_domains, selected_langs, show_anonymous ]: selector.change( update_table_long_doc, [ hidden_leaderboard_table_for_search, selected_domains, selected_langs, selected_rerankings, search_bar, show_anonymous, show_revision_and_timestamp ], leaderboard_table_long_doc, queue=True, ) # set metric listener selected_metric.change( update_metric_long_doc, [ selected_metric, selected_domains, selected_langs, selected_rerankings, search_bar, show_anonymous, show_revision_and_timestamp ], leaderboard_table_long_doc, queue=True ) select_noreranker_only_btn.click( clear_reranking_selections, outputs=selected_rerankings ) with gr.TabItem("🚀Submit here!", elem_id="submit-tab-table", id=2): with gr.Column(): with gr.Row(): gr.Markdown(EVALUATION_QUEUE_TEXT, elem_classes="markdown-text") with gr.Row(): gr.Markdown("## ✉️Submit your model here!", elem_classes="markdown-text") with gr.Row(): with gr.Column(): model_name = gr.Textbox(label="Retrieval Model name") with gr.Column(): model_url = gr.Textbox(label="Retrieval Model URL") with gr.Row(): with gr.Column(): reranking_model_name = gr.Textbox( label="Reranking Model name", info="Optional", value="NoReranker" ) with gr.Column(): reranking_model_url = gr.Textbox( label="Reranking Model URL", info="Optional", value="" ) with gr.Row(): with gr.Column(): benchmark_version = gr.Dropdown( ["AIR-Bench_24.04", ], value="AIR-Bench_24.04", interactive=True, label="AIR-Bench Version") with gr.Row(): upload_button = gr.UploadButton("Click to upload search results", file_count="single") with gr.Row(): file_output = gr.File() with gr.Row(): is_anonymous = gr.Checkbox( label="Nope. I want to submit anonymously 🥷", value=False, info="Do you want to shown on the leaderboard by default?") with gr.Row(): submit_button = gr.Button("Submit") with gr.Row(): submission_result = gr.Markdown() upload_button.upload( upload_file, [ upload_button, ], file_output) submit_button.click( submit_results, [ file_output, model_name, model_url, reranking_model_name, reranking_model_url, benchmark_version, is_anonymous ], submission_result, show_progress="hidden" ) with gr.TabItem("📝 About", elem_id="llm-benchmark-tab-table", id=3): gr.Markdown(BENCHMARKS_TEXT, elem_classes="markdown-text") scheduler = BackgroundScheduler() scheduler.add_job(restart_space, "interval", seconds=1800) scheduler.start() demo.queue(default_concurrency_limit=40).launch()