import gradio as gr import pandas as pd from apscheduler.schedulers.background import BackgroundScheduler from huggingface_hub import snapshot_download from gradio_space_ci import enable_space_ci from src.display.about import ( CITATION_BUTTON_LABEL, CITATION_BUTTON_TEXT, EVALUATION_QUEUE_TEXT, INTRODUCTION_TEXT, LLM_BENCHMARKS_TEXT, FAQ_TEXT, TITLE, ) from src.display.css_html_js import custom_css from src.display.utils import ( BENCHMARK_COLS, COLS, EVAL_COLS, EVAL_TYPES, NUMERIC_INTERVALS, TYPES, AutoEvalColumn, ModelType, fields, WeightType, Precision ) from src.envs import API, EVAL_REQUESTS_PATH, DYNAMIC_INFO_REPO, DYNAMIC_INFO_FILE_PATH, DYNAMIC_INFO_PATH, EVAL_RESULTS_PATH, H4_TOKEN, IS_PUBLIC, QUEUE_REPO, REPO_ID, RESULTS_REPO from src.populate import get_evaluation_queue_df, get_leaderboard_df 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.plots import ( create_metric_plot_obj, create_plot_df, create_scores_df, ) # Start ephemeral Spaces on PRs (see config in README.md) #enable_space_ci() def restart_space(): """ Restarts a Space instance specified by its repository ID. This function is used to restart a Space instance within the Hugging Face platform. It requires the repository ID and a valid API token for authentication. Parameters as env variables --------------------------- repo_id : str The ID of the repository associated with the Space instance to be restarted. token : str A valid API token with the necessary permissions to restart the Space. Returns ------- None This function does not return any value. It simply restarts the specified Space instance. Example ------- >>> restart_space(repo_id="example_repo_id", token="example_token") """ API.restart_space(repo_id=REPO_ID, token=H4_TOKEN) def init_space(): """ Initializes the Hugging Face Space environment. This function initializes the Hugging Face Space environment by performing the following steps: 1. Downloads evaluation requests, dynamic information, and evaluation results. 2. Processes the raw data into a leaderboard DataFrame. 3. Updates collections with the original DataFrame. 4. Creates a plot DataFrame for visualization. 5. Retrieves evaluation queue DataFrames. Returns ------- tuple A tuple containing the following elements: - leaderboard_df : pandas.DataFrame DataFrame containing the leaderboard data. - original_df : pandas.DataFrame Original DataFrame obtained from the evaluation results. - plot_df : pandas.DataFrame DataFrame suitable for creating plots. - finished_eval_queue_df : pandas.DataFrame DataFrame containing finished evaluation queue data. - running_eval_queue_df : pandas.DataFrame DataFrame containing running evaluation queue data. - pending_eval_queue_df : pandas.DataFrame DataFrame containing pending evaluation queue data. Example ------- >>> ( ... leaderboard_df, ... original_df, ... plot_df, ... finished_eval_queue_df, ... running_eval_queue_df, ... pending_eval_queue_df, ... ) = init_space() """ 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 ) except Exception: restart_space() try: print(DYNAMIC_INFO_PATH) snapshot_download( repo_id=DYNAMIC_INFO_REPO, local_dir=DYNAMIC_INFO_PATH, repo_type="dataset", tqdm_class=None, etag_timeout=30 ) 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 ) except Exception: restart_space() 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 ) update_collections(original_df.copy()) leaderboard_df = original_df.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 leaderboard_df, original_df, plot_df, finished_eval_queue_df, running_eval_queue_df, pending_eval_queue_df = init_space() # Searching and filtering def update_table( hidden_df: pd.DataFrame, columns: list, type_query: list, precision_query: str, size_query: list, hide_models: list, query: str, ): """ Updates a table DataFrame based on specified criteria. This function filters the input DataFrame based on specified criteria and returns a new DataFrame with selected columns. Parameters ---------- hidden_df : pandas.DataFrame The DataFrame to be filtered and updated. columns : list List of column names to be included in the updated DataFrame. type_query : list List of types to filter models. precision_query : str Precision value to filter models. size_query : list List of sizes to filter models. hide_models : list List of models to be hidden. query : str Query string to filter rows in the DataFrame. Returns ------- updated_df : pandas.DataFrame A DataFrame containing filtered and updated data based on the specified criteria. Example ------- >>> updated_df = update_table( ... hidden_df=original_df, ... columns=["Model", "Type", "Precision"], ... type_query=["type1", "type2"], ... precision_query="high", ... size_query=["large"], ... hide_models=["model1", "model2"], ... query="column1 > 0 and column2 == 'value'", ... ) """ 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, filtered_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 """ Loads a query parameter from a request object. It returns the query parameter value for the "search_bar" component and for a hidden component that triggers a reload only if the value has changed. Parameters ---------- request : gr.Request The request object containing query parameters. Returns ------- tuple A tuple containing two identical query parameter values: - query_search_bar : str The query parameter value for the "search_bar" component. - query_hidden : str The query parameter value for a hidden component that triggers a reload only if the value has changed. Example ------- >>> query_search_bar, query_hidden = load_query(request) """ 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: """ Searches a DataFrame for rows containing a specified query. This function filters the input DataFrame based on a specified query and returns a new DataFrame containing rows where the query matches any part of the specified column. Parameters ---------- df : pandas.DataFrame The DataFrame to be searched. query : str The query string to search for within the DataFrame. Returns ------- filtered_df : pandas.DataFrame A DataFrame containing rows where the query matches any part of the specified column. Example ------- >>> filtered_df = search_table(df=original_df, query="example_query") """ return df[(df[AutoEvalColumn.dummy.name].str.contains(query, case=False))] def select_columns(df: pd.DataFrame, columns: list) -> pd.DataFrame: """ Selects specified columns from a DataFrame. This function selects specified columns from the input DataFrame and returns a new DataFrame containing only those columns. Parameters ---------- df : pandas.DataFrame The DataFrame from which columns are to be selected. columns : list List of column names to be selected from the DataFrame. Returns ------- filtered_df : pandas.DataFrame A DataFrame containing only the specified columns. Example ------- >>> filtered_df = select_columns(df=original_df, columns=["column1", "column2", "column3"]) """ 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""" """ Filters DataFrame rows based on specified query strings. This function filters the input DataFrame based on specified query strings and returns a new DataFrame containing rows that match any of the queries. Parameters ---------- query : str The query string containing one or more search queries separated by semicolons (;). filtered_df : pandas.DataFrame The DataFrame to be filtered based on the queries. Returns ------- filtered_df : pandas.DataFrame A DataFrame containing rows that match any of the specified queries. Example ------- >>> filtered_df = filter_queries( ... query="query1; query2; query3", ... filtered_df=original_df, ... ) """ 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 filter_models( df: pd.DataFrame, type_query: list, size_query: list, precision_query: list, hide_models: list ) -> pd.DataFrame: """ Filters DataFrame rows based on specified criteria. This function filters the input DataFrame based on specified criteria such as model type, size, precision, and models to hide. Parameters ---------- df : pandas.DataFrame The DataFrame to be filtered. type_query : list List of tuples containing model types to include in the filtering. Each tuple consists of a model type abbreviation and its corresponding emoji. size_query : list List of size categories to include in the filtering. precision_query : list List of precision values to include in the filtering. hide_models : list List of model categories to hide from the DataFrame. Returns ------- filtered_df : pandas.DataFrame A DataFrame containing rows that meet the specified filtering criteria. Example ------- >>> filtered_df = filter_models( ... df=original_df, ... type_query=[("Type1", "🔥"), ("Type2", "⭐")], ... size_query=["Large", "Medium"], ... precision_query=["High", "Medium"], ... hide_models=["Private or deleted", "Contains a merge/moerge", "MoE", "Flagged"], ... ) """ # 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=["Private or deleted", "Contains a merge/moerge", "Flagged"], # 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("🏅 LLM Benchmark", elem_id="llm-benchmark-tab-table", id=0): 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, ) with gr.Row(): hide_models = gr.CheckboxGroup( label="Hide models", choices = ["Private or deleted", "Contains a merge/moerge", "Flagged", "MoE"], value=["Private or deleted", "Contains a merge/moerge", "Flagged"], interactive=True ) with gr.Column(min_width=320): #with gr.Box(elem_id="box-filter"): filter_columns_type = gr.CheckboxGroup( label="Model types", choices=[t.to_str() for t in ModelType], value=[t.to_str() for t in ModelType], interactive=True, elem_id="filter-columns-type", ) filter_columns_precision = gr.CheckboxGroup( label="Precision", choices=[i.value.name for i in Precision], value=[i.value.name for i in Precision], interactive=True, elem_id="filter-columns-precision", ) filter_columns_size = gr.CheckboxGroup( label="Model sizes (in billions of parameters)", choices=list(NUMERIC_INTERVALS.keys()), value=list(NUMERIC_INTERVALS.keys()), interactive=True, elem_id="filter-columns-size", ) leaderboard_table = gr.components.Dataframe( value=leaderboard_df[ [c.name for c in fields(AutoEvalColumn) if c.never_hidden] + shown_columns.value + [AutoEvalColumn.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], headers=COLS, datatype=TYPES, visible=False, ) search_bar.submit( update_table, [ hidden_leaderboard_table_for_search, shown_columns, filter_columns_type, filter_columns_precision, filter_columns_size, hide_models, 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, filter_columns_type, filter_columns_precision, filter_columns_size, hide_models, 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]: selector.change( update_table, [ hidden_leaderboard_table_for_search, shown_columns, filter_columns_type, filter_columns_precision, filter_columns_size, hide_models, search_bar, ], leaderboard_table, queue=True, ) with gr.TabItem("📈 Metrics through time", elem_id="llm-benchmark-tab-table", id=4): with gr.Row(): with gr.Column(): chart = create_metric_plot_obj( plot_df, [AutoEvalColumn.average.name], title="Average of Top Scores and Human Baseline Over Time (from last update)", ) gr.Plot(value=chart, min_width=500) with gr.Column(): chart = create_metric_plot_obj( plot_df, BENCHMARK_COLS, title="Top Scores and Human Baseline Over Time (from last update)", ) gr.Plot(value=chart, min_width=500) with gr.TabItem("📝 About", elem_id="llm-benchmark-tab-table", id=2): gr.Markdown(LLM_BENCHMARKS_TEXT, elem_classes="markdown-text") gr.Markdown(FAQ_TEXT, elem_classes="markdown-text") with gr.TabItem("🚀 Submit here! ", elem_id="llm-benchmark-tab-table", id=3): with gr.Column(): with gr.Row(): gr.Markdown(EVALUATION_QUEUE_TEXT, elem_classes="markdown-text") with gr.Column(): with gr.Accordion( f"✅ Finished Evaluations ({len(finished_eval_queue_df)})", open=False, ): with gr.Row(): finished_eval_table = gr.components.Dataframe( value=finished_eval_queue_df, headers=EVAL_COLS, datatype=EVAL_TYPES, row_count=5, ) with gr.Accordion( f"🔄 Running Evaluation Queue ({len(running_eval_queue_df)})", open=False, ): with gr.Row(): running_eval_table = gr.components.Dataframe( value=running_eval_queue_df, headers=EVAL_COLS, datatype=EVAL_TYPES, row_count=5, ) with gr.Accordion( f"⏳ Pending Evaluation Queue ({len(pending_eval_queue_df)})", open=False, ): with gr.Row(): pending_eval_table = gr.components.Dataframe( value=pending_eval_queue_df, headers=EVAL_COLS, datatype=EVAL_TYPES, row_count=5, ) with gr.Row(): gr.Markdown("# ✉️✨ Submit your model here!", elem_classes="markdown-text") with gr.Row(): with gr.Column(): model_name_textbox = gr.Textbox(label="Model name") revision_name_textbox = gr.Textbox(label="Revision commit", placeholder="main") private = gr.Checkbox(False, label="Private", visible=not IS_PUBLIC) model_type = gr.Dropdown( choices=[t.to_str(" : ") for t in ModelType if t != ModelType.Unknown], label="Model type", multiselect=False, value=ModelType.FT.to_str(" : "), interactive=True, ) with gr.Column(): precision = gr.Dropdown( choices=[i.value.name for i in Precision if i != Precision.Unknown], label="Precision", multiselect=False, value="float16", interactive=True, ) weight_type = gr.Dropdown( choices=[i.value.name for i in WeightType], label="Weights type", multiselect=False, value="Original", interactive=True, ) base_model_name_textbox = gr.Textbox(label="Base model (for delta or adapter weights)") submit_button = gr.Button("Submit Eval") submission_result = gr.Markdown() submit_button.click( add_new_eval, [ model_name_textbox, base_model_name_textbox, revision_name_textbox, precision, private, weight_type, model_type, ], submission_result, ) 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=10800) # restarted every 3h scheduler.add_job(update_dynamic_files, "cron", minute=30) # launched every hour on the hour scheduler.start() demo.queue(default_concurrency_limit=40).launch()