import os import json import gradio as gr 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, ) from src.display.css_html_js import custom_css from src.display.utils import ( BENCHMARK_COLS, COLS, EVAL_COLS, EVAL_TYPES, AutoEvalColumn, ModelType, fields, WeightType, Precision, AddSpecialTokens, NumFewShots, NUMERIC_INTERVALS, TYPES, ) from src.envs import API, EVAL_REQUESTS_PATH, EVAL_RESULTS_PATH, QUEUE_REPO, REPO_ID, RESULTS_REPO, TOKEN from src.populate import get_evaluation_queue_df, get_leaderboard_df from src.submission.submit import add_new_eval def restart_space(): API.restart_space(repo_id=REPO_ID) ### 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() LEADERBOARD_DF = get_leaderboard_df(EVAL_RESULTS_PATH, EVAL_REQUESTS_PATH, COLS, BENCHMARK_COLS) original_df = LEADERBOARD_DF leaderboard_df = original_df.copy() ( finished_eval_queue_df, running_eval_queue_df, pending_eval_queue_df, failed_eval_queue_df, ) = get_evaluation_queue_df(EVAL_REQUESTS_PATH, EVAL_COLS) # Searching and filtering def update_table( hidden_df: pd.DataFrame, columns: list, type_query: list, precision_query: str, size_query: list, add_special_tokens_query: list, num_few_shots_query: list, show_deleted: bool, show_merges: bool, show_flagged: bool, query: str, ): print(f"Update table called with: type_query={type_query}, precision_query={precision_query}, size_query={size_query}") print(f"hidden_df shape before filtering: {hidden_df.shape}") filtered_df = filter_models(hidden_df, type_query, size_query, precision_query, add_special_tokens_query, num_few_shots_query, show_deleted, show_merges, show_flagged) print(f"filtered_df shape after filter_models: {filtered_df.shape}") filtered_df = filter_queries(query, filtered_df) print(f"filtered_df shape after filter_queries: {filtered_df.shape}") print(f"Filter applied: query={query}, columns={columns}, type_query={type_query}, precision_query={precision_query}") print("Filtered dataframe head:") print(filtered_df.head()) df = select_columns(filtered_df, columns) print(f"Final df shape: {df.shape}") print("Final dataframe head:") print(df.head()) 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 = [ # 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]# + [AutoEvalColumn.dummy.name] # ] # return filtered_df def select_columns(df: pd.DataFrame, columns: list) -> pd.DataFrame: always_here_cols = [ AutoEvalColumn.model_type_symbol.name, AutoEvalColumn.model.name, ] selected_cols = list(dict.fromkeys(always_here_cols + [c for c in COLS if c in df.columns and c in columns])) return df[selected_cols] 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 filter_models( df: pd.DataFrame, type_query: list, size_query: list, precision_query: list, add_special_tokens_query: list, num_few_shots_query: list, show_deleted: bool, show_merges: bool, show_flagged: bool ) -> pd.DataFrame: print(f"Initial df shape: {df.shape}") print(f"Initial df content:\n{df}") filtered_df = df # Model Type フィルタリング type_emoji = [t.split()[0] for t in type_query] filtered_df = filtered_df[filtered_df['T'].isin(type_emoji)] print(f"After type filter: {filtered_df.shape}") # Precision フィルタリング filtered_df = filtered_df[filtered_df['Precision'].isin(precision_query + ['Unknown', '?'])] print(f"After precision filter: {filtered_df.shape}") # Model Size フィルタリング if 'Unknown' in size_query: size_mask = filtered_df['#Params (B)'].isna() | (filtered_df['#Params (B)'] == 0) else: size_mask = filtered_df['#Params (B)'].apply(lambda x: any(x in NUMERIC_INTERVALS[s] for s in size_query if s != 'Unknown')) filtered_df = filtered_df[size_mask] print(f"After size filter: {filtered_df.shape}") # Add Special Tokens フィルタリング filtered_df = filtered_df[filtered_df['Add Special Tokens'].isin(add_special_tokens_query + ['Unknown', '?'])] print(f"After add_special_tokens filter: {filtered_df.shape}") # Num Few Shots フィルタリング filtered_df = filtered_df[filtered_df['Few-shot'].astype(str).isin([str(x) for x in num_few_shots_query] + ['Unknown', '?'])] print(f"After num_few_shots filter: {filtered_df.shape}") # Show deleted models フィルタリング if not show_deleted: filtered_df = filtered_df[filtered_df['Available on the hub'] == True] print(f"After show_deleted filter: {filtered_df.shape}") print("Filtered dataframe head:") print(filtered_df.head()) return filtered_df leaderboard_df = filter_models(leaderboard_df, [t.to_str(" : ") for t in ModelType], list(NUMERIC_INTERVALS.keys()), [i.value.name for i in Precision], [i.value.name for i in AddSpecialTokens], [i.value.name for i in NumFewShots], False, False, False) 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(): deleted_models_visibility = gr.Checkbox( value=False, label="Show private/deleted models", interactive=True ) merged_models_visibility = gr.Checkbox( value=False, label="Show merges", interactive=True ) flagged_models_visibility = gr.Checkbox( value=False, label="Show flagged models", 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", ) filter_columns_add_special_tokens = gr.CheckboxGroup( label="Add Special Tokens", choices=[i.value.name for i in AddSpecialTokens], value=[i.value.name for i in AddSpecialTokens], interactive=True, elem_id="filter-columns-add-special-tokens", ) filter_columns_num_few_shots = gr.CheckboxGroup( label="Num Few Shots", choices=[i.value.name for i in NumFewShots], value=[i.value.name for i in NumFewShots], interactive=True, elem_id="filter-columns-num-few-shots", ) leaderboard_df_filtered = filter_models(leaderboard_df, [t.to_str(" : ") for t in ModelType], list(NUMERIC_INTERVALS.keys()), [i.value.name for i in Precision], [i.value.name for i in AddSpecialTokens], [i.value.name for i in NumFewShots], False, False, False) initial_columns = [c.name for c in fields(AutoEvalColumn) if c.never_hidden or c.displayed_by_default] leaderboard_df_filtered = select_columns(leaderboard_df_filtered, initial_columns) # leaderboard_table = gr.components.Dataframe( # value=leaderboard_df_filtered, # 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, # ) print("Columns in leaderboard_df_filtered:", leaderboard_df_filtered.columns) datatype_dict = {col: "markdown" if col == "Model" else "str" for col in leaderboard_df_filtered.columns} leaderboard_table = gr.components.Dataframe( value=leaderboard_df_filtered.to_dict('records'), # DataFrame を辞書のリストに変換 headers=list(leaderboard_df_filtered.columns), datatype=datatype_dict, 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[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, filter_columns_add_special_tokens, filter_columns_num_few_shots, deleted_models_visibility, merged_models_visibility, flagged_models_visibility, search_bar, ], leaderboard_table, ) # Define a hidden component that will trigger a reload only if a query parameter has be 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, filter_columns_add_special_tokens, filter_columns_num_few_shots, deleted_models_visibility, merged_models_visibility, flagged_models_visibility, 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, filter_columns_add_special_tokens, filter_columns_num_few_shots, deleted_models_visibility, merged_models_visibility, flagged_models_visibility]: selector.change( update_table, [ hidden_leaderboard_table_for_search, shown_columns, filter_columns_type, filter_columns_precision, filter_columns_size, filter_columns_add_special_tokens, filter_columns_num_few_shots, deleted_models_visibility, merged_models_visibility, flagged_models_visibility, search_bar, ], leaderboard_table, queue=True, ) with gr.TabItem("📝 About", elem_id="llm-benchmark-tab-table", id=2): gr.Markdown(LLM_BENCHMARKS_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.Accordion( f"❎ Failed Evaluation Queue ({len(failed_eval_queue_df)})", open=False, ): with gr.Row(): failed_eval_table = gr.components.Dataframe( value=failed_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") model_type = gr.Dropdown( choices=[t.to_str(" : ") for t in ModelType if t != ModelType.Unknown], label="Model type", multiselect=False, value=None, 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)") add_special_tokens = gr.Dropdown( choices=[i.value.name for i in AddSpecialTokens if i != AddSpecialTokens.Unknown], label="AddSpecialTokens", multiselect=False, value="False", interactive=True, ) 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, weight_type, model_type, add_special_tokens, ], 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=1800) scheduler.start() demo.queue(default_concurrency_limit=40).launch()