import gradio as gr from gradio_leaderboard import Leaderboard, ColumnFilter, SelectColumns import pandas as pd from apscheduler.schedulers.background import BackgroundScheduler from huggingface_hub import snapshot_download from src.about import ( CITATION_BUTTON_LABEL, CITATION_BUTTON_TEXT, EVALUATION_QUEUE_TEXT, INTRODUCTION_TEXT, LLM_BENCHMARKS_TEXT, TITLE, EVALUATION_METRIC_TEXT, ) from src.display.css_html_js import custom_css from src.display.utils import ( BENCHMARK_COLS, COLS, EVAL_COLS, EVAL_TYPES, TYPES, AutoEvalColumn, ModelType, fields, WeightType, Precision, NUMERIC_INTERVALS, QUOTACOLS, QUOTATYPES, AutoEvalColumnQuota, BENCHMARK_QUOTACOLS ) from src.envs import API, EVAL_REQUESTS_PATH, EVAL_RESULTS_PATH, DYNAMIC_INFO_REPO, DYNAMIC_INFO_FILE_PATH, DYNAMIC_INFO_PATH, IS_PUBLIC, QUEUE_REPO, REPO_ID, RESULTS_REPO, TOKEN from src.populate import get_evaluation_queue_df, get_leaderboard_df, get_leaderboard_df_quota from src.submission.submit import add_new_eval from src.scripts.update_all_request_files import update_dynamic_files from src.tools.collections import update_collections from src.tools.datastatics import get_statics #from src.tools.plots import ( # create_plot_df, # create_scores_df, #) def restart_space(): API.restart_space(repo_id=REPO_ID, token=TOKEN) def init_space(): print("begin init space") ### Space initialisation try: print(EVAL_REQUESTS_PATH) snapshot_download( repo_id=QUEUE_REPO, local_dir=EVAL_REQUESTS_PATH, repo_type="dataset", tqdm_class=None, etag_timeout=30, token=TOKEN ) except Exception: restart_space() try: print(EVAL_RESULTS_PATH) snapshot_download( repo_id=RESULTS_REPO, local_dir=EVAL_RESULTS_PATH, repo_type="dataset", tqdm_class=None, etag_timeout=30, token=TOKEN ) except Exception: restart_space() print("QUOTACOLS+COLS",list(set(QUOTACOLS+COLS))) raw_data, original_df = get_leaderboard_df( results_path=EVAL_RESULTS_PATH, requests_path=EVAL_REQUESTS_PATH, dynamic_path=DYNAMIC_INFO_FILE_PATH, #cols=COLS, #benchmark_cols=BENCHMARK_COLS, cols=list(set(QUOTACOLS+COLS)), benchmark_cols=list(set(BENCHMARK_QUOTACOLS+BENCHMARK_COLS)) ) #update_collections(original_df.copy()) leaderboard_df = original_df.copy() raw_data_quota, original_df_quota = get_leaderboard_df( results_path=EVAL_RESULTS_PATH, requests_path=EVAL_REQUESTS_PATH, dynamic_path=DYNAMIC_INFO_FILE_PATH, cols=list(set(QUOTACOLS+COLS)), benchmark_cols=list(set(BENCHMARK_QUOTACOLS+BENCHMARK_COLS)) ) #update_collections(original_df.copy()) leaderboard_df_quota = original_df_quota.copy() #plot_df = create_plot_df(create_scores_df(raw_data)) ( finished_eval_queue_df, running_eval_queue_df, pending_eval_queue_df, ) = get_evaluation_queue_df(EVAL_REQUESTS_PATH, EVAL_COLS) #return leaderboard_df, original_df, plot_df, finished_eval_queue_df, running_eval_queue_df, pending_eval_queue_df return leaderboard_df, original_df, leaderboard_df_quota, original_df_quota,finished_eval_queue_df, running_eval_queue_df, pending_eval_queue_df leaderboard_df, original_df, leaderboard_df_quota, original_df_quota, finished_eval_queue_df, running_eval_queue_df, pending_eval_queue_df = init_space() # Searching and filtering #type_query: list, #precision_query: str, #size_query: list, #hide_models: list, def update_table( hidden_df: pd.DataFrame, columns: list, query: str, ): print("query", query) #filtered_df = filter_models(df=hidden_df, type_query=type_query, size_query=size_query, precision_query=precision_query, hide_models=hide_models) filtered_df = filter_queries(query, hidden_df) df = select_columns(filtered_df, columns) return df def load_query(request: gr.Request): # triggered only once at startup => read query parameter if it exists query = request.query_params.get("query") or "" return query, query # return one for the "search_bar", one for a hidden component that triggers a reload only if value has changed def search_table(df: pd.DataFrame, query: str) -> pd.DataFrame: return df[(df[AutoEvalColumn.dummy.name].str.contains(query, case=False))] def select_columns(df: pd.DataFrame, columns: list) -> pd.DataFrame: always_here_cols = [c.name for c in fields(AutoEvalColumn) if c.never_hidden] dummy_col = [AutoEvalColumn.dummy.name] #AutoEvalColumn.model_type_symbol.name, #AutoEvalColumn.model.name, # We use COLS to maintain sorting filtered_df = df[ always_here_cols + [c for c in COLS if c in df.columns and c in columns] + dummy_col ] return filtered_df def filter_queries(query: str, filtered_df: pd.DataFrame): """Added by Abishek""" final_df = [] if query != "": queries = [q.strip() for q in query.split(";")] for _q in queries: _q = _q.strip() if _q != "": temp_filtered_df = search_table(filtered_df, _q) if len(temp_filtered_df) > 0: final_df.append(temp_filtered_df) if len(final_df) > 0: filtered_df = pd.concat(final_df) filtered_df = filtered_df.drop_duplicates( subset=[AutoEvalColumn.model.name, AutoEvalColumn.precision.name, AutoEvalColumn.revision.name] ) return filtered_df def update_table_q( hidden_df: pd.DataFrame, columns: list, query: str, ): filtered_df = filter_queriesq(query, hidden_df) df = select_columnsq(filtered_df, columns) return df def search_tableq(df: pd.DataFrame, query: str) -> pd.DataFrame: return df[(df[AutoEvalColumnQuota.dummy.name].str.contains(query, case=False))] def select_columnsq(df: pd.DataFrame, columns: list) -> pd.DataFrame: always_here_cols = [c.name for c in fields(AutoEvalColumnQuota) if c.never_hidden] dummy_col = [AutoEvalColumnQuota.dummy.name] # We use COLS to maintain sorting filtered_df = df[ always_here_cols + [c for c in QUOTACOLS if c in df.columns and c in columns] + dummy_col ] return filtered_df def filter_queriesq(query: str, filtered_df: pd.DataFrame): """Added by Abishek""" final_df = [] if query != "": queries = [q.strip() for q in query.split(";")] for _q in queries: _q = _q.strip() if _q != "": temp_filtered_df = search_tableq(filtered_df, _q) if len(temp_filtered_df) > 0: final_df.append(temp_filtered_df) if len(final_df) > 0: filtered_df = pd.concat(final_df) filtered_df = filtered_df.drop_duplicates( subset=[AutoEvalColumnQuota.model.name, AutoEvalColumnQuota.precision.name, AutoEvalColumnQuota.revision.name] ) return filtered_df def filter_models( df: pd.DataFrame, type_query: list, size_query: list, precision_query: list, hide_models: list ) -> pd.DataFrame: # Show all models if "Private or deleted" in hide_models: filtered_df = df[df[AutoEvalColumn.still_on_hub.name] == True] else: filtered_df = df if "Contains a merge/moerge" in hide_models: filtered_df = filtered_df[filtered_df[AutoEvalColumn.merged.name] == False] if "MoE" in hide_models: filtered_df = filtered_df[filtered_df[AutoEvalColumn.moe.name] == False] if "Flagged" in hide_models: filtered_df = filtered_df[filtered_df[AutoEvalColumn.flagged.name] == False] type_emoji = [t[0] for t in type_query] filtered_df = filtered_df.loc[df[AutoEvalColumn.model_type_symbol.name].isin(type_emoji)] filtered_df = filtered_df.loc[df[AutoEvalColumn.precision.name].isin(precision_query + ["None"])] numeric_interval = pd.IntervalIndex(sorted([NUMERIC_INTERVALS[s] for s in size_query])) params_column = pd.to_numeric(df[AutoEvalColumn.params.name], errors="coerce") mask = params_column.apply(lambda x: any(numeric_interval.contains(x))) filtered_df = filtered_df.loc[mask] return filtered_df leaderboard_df = filter_models( df=leaderboard_df, type_query=[t.to_str(" : ") for t in ModelType], size_query=list(NUMERIC_INTERVALS.keys()), precision_query=[i.value.name for i in Precision], hide_models=[], # Deleted, merges, flagged, MoEs ) leaderboard_df_quota = filter_models( df=leaderboard_df_quota, type_query=[t.to_str(" : ") for t in ModelType], size_query=list(NUMERIC_INTERVALS.keys()), precision_query=[i.value.name for i in Precision], hide_models=[], # Deleted, merges, flagged, MoEs ) 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("🏅 EmbodiedVerse-Open", elem_id="vlm-benchmark-tab-table", id=0): #leaderboard = init_leaderboard(LEADERBOARD_DF) with gr.Row(): with gr.Column(): with gr.Row(): search_bar = gr.Textbox( placeholder=" 🔍 Search for your model (separate multiple queries with `;`) and press ENTER...", show_label=False, elem_id="search-bar", ) with gr.Row(): shown_columns = gr.CheckboxGroup( choices=[ c.name for c in fields(AutoEvalColumn) if not c.hidden and not c.never_hidden and not c.dummy ], value=[ c.name for c in fields(AutoEvalColumn) if c.displayed_by_default and not c.hidden and not c.never_hidden ], label="Select columns to show", elem_id="column-select", interactive=True, ) leaderboard_table = gr.components.Dataframe( value=leaderboard_df[ [c.name for c in fields(AutoEvalColumn) if c.never_hidden] + shown_columns.value + [AutoEvalColumnQuota.dummy.name] ], headers=[c.name for c in fields(AutoEvalColumn) if c.never_hidden] + shown_columns.value, datatype=TYPES, elem_id="leaderboard-table", interactive=False, visible=True, #column_widths=["2%", "33%"] ) # Dummy leaderboard for handling the case when the user uses backspace key hidden_leaderboard_table_for_search = gr.components.Dataframe( value=original_df[COLS], #value=leaderboard_df[COLS], headers=COLS, datatype=TYPES, visible=False, ) search_bar.submit( update_table, [ hidden_leaderboard_table_for_search, shown_columns, search_bar, ], leaderboard_table, ) # Define a hidden component that will trigger a reload only if a query parameter has been set hidden_search_bar = gr.Textbox(value="", visible=False) hidden_search_bar.change( update_table, [ hidden_leaderboard_table_for_search, shown_columns, search_bar, ], leaderboard_table, ) # Check query parameter once at startup and update search bar + hidden component demo.load(load_query, inputs=[], outputs=[search_bar, hidden_search_bar]) #for selector in [shown_columns, filter_columns_type, filter_columns_precision, filter_columns_size, hide_models]: for selector in [shown_columns]:#, filter_columns_type, filter_columns_precision, filter_columns_size]:#, hide_models]: selector.change( update_table, [ hidden_leaderboard_table_for_search, shown_columns, search_bar, ], leaderboard_table, queue=True, ) with gr.TabItem("🏅 EmbodiedVerse-Open-Sampled", elem_id="vlm-quota-benchmark-tab-table", id=1): #leaderboard = init_leaderboard(LEADERBOARD_DF) with gr.Row(): with gr.Column(): with gr.Row(): search_bar = gr.Textbox( placeholder=" 🔍 Search for your model (separate multiple queries with `;`) and press ENTER...", show_label=False, elem_id="search-bar", ) with gr.Row(): shown_columns = gr.CheckboxGroup( choices=[ c.name for c in fields(AutoEvalColumnQuota) if not c.hidden and not c.never_hidden and not c.dummy ], value=[ c.name for c in fields(AutoEvalColumnQuota) if c.displayed_by_default and not c.hidden and not c.never_hidden ], label="Select columns to show", elem_id="column-select", interactive=True, ) leaderboard_table = gr.components.Dataframe( value=leaderboard_df_quota[ [c.name for c in fields(AutoEvalColumnQuota) if c.never_hidden] + shown_columns.value + [AutoEvalColumnQuota.dummy.name] ], headers=[c.name for c in fields(AutoEvalColumnQuota) if c.never_hidden] + shown_columns.value, datatype=QUOTATYPES, elem_id="leaderboard-table", interactive=False, visible=True, #column_widths=["2%", "33%"] ) # Dummy leaderboard for handling the case when the user uses backspace key hidden_leaderboard_table_for_search = gr.components.Dataframe( value=original_df_quota[QUOTACOLS], headers=QUOTACOLS, datatype=QUOTATYPES, visible=False, ) search_bar.submit( update_table_q, [ hidden_leaderboard_table_for_search, shown_columns, search_bar, ], leaderboard_table, ) # Define a hidden component that will trigger a reload only if a query parameter has been set hidden_search_bar = gr.Textbox(value="", visible=False) hidden_search_bar.change( update_table_q, [ hidden_leaderboard_table_for_search, shown_columns, search_bar, ], leaderboard_table, ) # Check query parameter once at startup and update search bar + hidden component demo.load(load_query, inputs=[], outputs=[search_bar, hidden_search_bar]) #for selector in [shown_columns, filter_columns_type, filter_columns_precision, filter_columns_size, hide_models]: for selector in [shown_columns]:#, filter_columns_type, filter_columns_precision, filter_columns_size]: #, hide_models]: selector.change( update_table_q, [ hidden_leaderboard_table_for_search, shown_columns, search_bar, ], leaderboard_table, queue=True, ) gr.Markdown(EVALUATION_METRIC_TEXT, elem_classes="markdown-text") with gr.TabItem("📝 About", elem_id="llm-benchmark-tab-table", id=2): gr.Markdown(LLM_BENCHMARKS_TEXT, elem_classes="markdown-text") 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, ) scheduler = BackgroundScheduler() scheduler.add_job(restart_space, "interval", seconds=1800) scheduler.add_job(update_dynamic_files, "cron", minute=30) # launched every hour on the hour scheduler.add_job(get_statics, 'cron', hour=12, minute=15, timezone='Asia/Shanghai') # 添加定时任务，每天0:30执行一次 scheduler.start() demo.queue(default_concurrency_limit=40).launch()