import subprocess 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, NUMERIC_INTERVALS, TYPES, AutoEvalColumn, ModelType, fields, WeightType, Precision ) 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 from PIL import Image from dummydatagen import dummy_data_for_plot, create_metric_plot_obj_1, dummydf import copy def load_data(data_path): columns = ['Unlearned_Methods','Source', 'Diffusion_Models','Pre-ASR', 'Post-ASR','FID'] columns_sorted = ['Unlearned_Methods','Source', 'Diffusion_Models','Pre-ASR', 'Post-ASR','FID'] df = pd.read_csv(data_path).dropna() df['Post-ASR'] = df['Post-ASR'].round(0) # rank according to the Score column df = df.sort_values(by='Post-ASR', ascending=False) # reorder the columns df = df[columns_sorted] return df def restart_space(): API.restart_space(repo_id=REPO_ID) # 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() # raw_data, original_df = get_leaderboard_df(EVAL_RESULTS_PATH, EVAL_REQUESTS_PATH, COLS, BENCHMARK_COLS) # leaderboard_df = original_df.copy() # ( # finished_eval_queue_df, # running_eval_queue_df, # pending_eval_queue_df, # ) = get_evaluation_queue_df(EVAL_REQUESTS_PATH, EVAL_COLS) csv_path='./assets/object_parachute.csv' df_results = load_data(csv_path) methods = list(set(df_results['Unlearned_Methods'])) all_columns = ['Unlearned_Methods','Source', 'Diffusion_Models','Pre-ASR', 'Post-ASR','FID'] show_columns = ['Unlearned_Methods','Source', 'Diffusion_Models','Pre-ASR', 'Post-ASR','FID'] TYPES = ['str', 'markdown', 'str', 'number', 'number', 'number'] files = ['church','garbage','parachute','tench', 'vangogh', 'nidity', 'violence','illegal_activity'] df_results_init = df_results.copy()[show_columns] def update_table( hidden_df: pd.DataFrame, model1_column: list, #type_query: list, #open_query: list, # precision_query: str, # size_query: list, # show_deleted: bool, query: str, ): # filtered_df = filter_models(hidden_df, type_query, size_query, precision_query, show_deleted) # filtered_df = filter_queries(query, filtered_df) # df = select_columns(filtered_df, columns) filtered_df = hidden_df.copy() # print(open_query) # filtered_df = filtered_df[filtered_df['Unlearned_Methods'].isin(open_query)] # map_open = {'open': 'Yes', 'closed': 'No'} # filtered_df = filtered_df[filtered_df['Open?'].isin([map_open[o] for o in open_query])] filtered_df=select_columns(filtered_df,model1_column) filtered_df = filter_queries(query, filtered_df) # map_open = {'SD V1.4', 'SD V1.5', 'SD V2.0'} # filtered_df = filtered_df[filtered_df["Diffusion_Models"].isin([o for o in open_query])] # filtered_df = filtered_df[[map_columns[k] for k in columns]] # deduplication # df = df.drop_duplicates(subset=["Model"]) df = filtered_df.drop_duplicates() # df = df[show_columns] return df def search_table(df: pd.DataFrame, query: str) -> pd.DataFrame: return df[(df['Unlearned_Methods'].str.contains(query, case=False))] def filter_queries(query: str, filtered_df: pd.DataFrame) -> pd.DataFrame: 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) return filtered_df def search_table_model(df: pd.DataFrame, query: str) -> pd.DataFrame: return df[(df['Diffusion_Models'].str.contains(query, case=False))] def filter_queries_model(query: str, filtered_df: pd.DataFrame) -> pd.DataFrame: final_df = [] # if query != "": # queries = [q.strip() for q in query.split(";")] for _q in query: print(_q) if _q != "": temp_filtered_df = search_table_model(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) return filtered_df def select_columns(df: pd.DataFrame, columns_1: list) -> pd.DataFrame: always_here_cols = ['Unlearned_Methods','Source', 'Diffusion_Models'] # We use COLS to maintain sorting all_columns =['Pre-ASR','Post-ASR','FID'] if (len(columns_1)) == 0: filtered_df = df[ always_here_cols + [c for c in all_columns if c in df.columns] ] else: filtered_df = df[ always_here_cols + [c for c in all_columns if c in df.columns and (c in columns_1) ] ] return filtered_df demo = gr.Blocks(css=custom_css) with demo: gr.HTML(TITLE) gr.Markdown(INTRODUCTION_TEXT, elem_classes="markdown-text") gr.Markdown(EVALUATION_QUEUE_TEXT,elem_classes="eval-text") with gr.Tabs(elem_classes="tab-buttons") as tabs: with gr.TabItem("UnlearnDiffAtk Benchmark", elem_id="UnlearnDiffAtk-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(): model1_column = gr.CheckboxGroup( label="Evaluation Metrics", choices=['Pre-ASR', 'Post-ASR','FID'], interactive=True, elem_id="column-select", ) # with gr.Column(min_width=320): # with gr.Row(): # shown_columns_1 = gr.CheckboxGroup( # choices=["Church","Parachute","Tench", "Garbage Truck"], # label="Undersirable Objects", # elem_id="column-object", # interactive=True, # ) # with gr.Row(): # shown_columns_2 = gr.CheckboxGroup( # choices=["Van Gogh"], # label="Undersirable Styles", # elem_id="column-style", # interactive=True, # ) # with gr.Row(): # shown_columns_3 = gr.CheckboxGroup( # choices=["Violence","Illegal Activity","Nudity"], # label="Undersirable Concepts (Outputs that may be offensive in nature)", # elem_id="column-select", # interactive=True, # ) # with gr.Row(): # shown_columns_4 = gr.Slider( # 1, 100, value=40, # step=1, label="Attacking Steps", info="Choose between 1 and 100", # interactive=True,) for i in range(len(files)): if files[i] == "church": name = "### Unlearned Objects "+" Church" elif file[i] == 'garbage': name = "### Unlearned Objects "+" Garbage" elif file[i] == 'tench': name = "### Unlearned Objects "+" Tench" elif file[i] == 'parachute': name = "### Unlearned Objects "+" Parachute" elif file[i] == 'vangogh': name = "### Unlearned Stype "+" Van Gogh" elif file[i] == 'nudity': name = "### Unlearned Concepts "+" Nudity" elif file[i] == 'violence': name = "### Unlearned Concepts "+" Violence" elif file[i] == 'ill': name = "### Unlearned Concepts "+" Illgal Activity" gr.Markdown(name) gr.Markdown("### Unlearned Concepts Parachute") leaderboard_table = gr.components.Dataframe( value = df_results, datatype = TYPES, elem_id = "leaderboard-table", interactive = False, visible=True, ) hidden_leaderboard_table_for_search = gr.components.Dataframe( value=df_results_init, # elem_id="leaderboard-table", interactive=False, visible=False, ) search_bar.submit( update_table, [ # df_avg, hidden_leaderboard_table_for_search, model1_column, # shown_columns, #type_query, #open_query, # filter_columns_type, # filter_columns_precision, # filter_columns_size, # deleted_models_visibility, search_bar, ], leaderboard_table, ) #for selector in [type_query, open_query]: for selector in [model1_column]: selector.change( update_table, [ # df_avg, hidden_leaderboard_table_for_search, model1_column, # shown_columns, #type_query, #open_query, # filter_columns_type, # filter_columns_precision, # filter_columns_size, # deleted_models_visibility, search_bar, ], leaderboard_table, ) gr.Markdown("### Unlearned Concepts Parachute") leaderboard_table = gr.components.Dataframe( value = df_results, datatype = TYPES, elem_id = "leaderboard-table", interactive = False, visible=True, ) hidden_leaderboard_table_for_search = gr.components.Dataframe( value=df_results_init, # elem_id="leaderboard-table", interactive=False, visible=False, ) search_bar.submit( update_table, [ # df_avg, hidden_leaderboard_table_for_search, model1_column, # shown_columns, #type_query, #open_query, # filter_columns_type, # filter_columns_precision, # filter_columns_size, # deleted_models_visibility, search_bar, ], leaderboard_table, ) #for selector in [type_query, open_query]: for selector in [model1_column]: selector.change( update_table, [ # df_avg, hidden_leaderboard_table_for_search, model1_column, # shown_columns, #type_query, #open_query, # filter_columns_type, # filter_columns_precision, # filter_columns_size, # deleted_models_visibility, search_bar, ], leaderboard_table, ) with gr.Row(): with gr.Accordion("📙 Citation", open=True): citation_button = gr.Textbox( value=CITATION_BUTTON_TEXT, label=CITATION_BUTTON_LABEL, lines=10, elem_id="citation-button", show_copy_button=True, ) scheduler = BackgroundScheduler() scheduler.add_job(restart_space, "interval", seconds=1800) scheduler.start() demo.queue().launch(share=True)