import json import gradio as gr import pandas as pd from apscheduler.schedulers.background import BackgroundScheduler from huggingface_hub import snapshot_download from src.display.about import ( CITATION_BUTTON_LABEL, CITATION_BUTTON_TEXT, EVALUATION_QUEUE_TEXT, INTRODUCTION_TEXT, LLM_BENCHMARKS_TEXT, TITLE, TABLE_DESC, ) from src.display.css_html_js import custom_css from src.display.formatting import styled_error, styled_message, styled_warning, model_hyperlink from src.display.utils import ( BENCHMARK_COLS, COLS, EVAL_COLS, EVAL_TYPES, NUMERIC_INTERVALS, TYPES, TYPES_LITE, AutoEvalColumn, ModelType, fields, WeightType, Precision, ) from src.envs import API, EVAL_REQUESTS_PATH, EVAL_RESULTS_PATH, TOKEN, 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 captcha.image import ImageCaptcha from PIL import Image import random, string import matplotlib.pyplot as plt original_df = None leaderboard_df = None def restart_space(): API.restart_space(repo_id=REPO_ID, token=TOKEN) def add_model_hyperlinks(row): if row["Model URL"] is None or row["Model URL"] == "": return row["Model"] else: return model_hyperlink(row["Model URL"], row["Model"]) def download_data(): global original_df global leaderboard_df try: print(EVAL_REQUESTS_PATH,QUEUE_REPO) 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(EVAL_RESULTS_PATH, RESULTS_REPO) snapshot_download( repo_id=RESULTS_REPO, local_dir=EVAL_RESULTS_PATH, repo_type="dataset", tqdm_class=None, etag_timeout=30 ) except Exception: restart_space() _, original_df = get_leaderboard_df(EVAL_RESULTS_PATH, EVAL_REQUESTS_PATH, COLS, BENCHMARK_COLS) leaderboard_df = original_df.copy() leaderboard_df["Model"] = leaderboard_df.apply(add_model_hyperlinks, axis=1) leaderboard_df.sort_values(by=["Aggregate Score"], ascending=False, inplace=True) download_data() """ ( finished_eval_queue_df, running_eval_queue_df, pending_eval_queue_df, ) = get_evaluation_queue_df(EVAL_REQUESTS_PATH, EVAL_COLS) """ # Searching and filtering def update_table( hidden_df: pd.DataFrame, columns: list, query: str, ): columns += " " # The dataframe does not display the last column - BUG in gradio? #filtered_df = filter_models(hidden_df, type_query, size_query, precision_query, show_deleted) filtered_df = filter_queries(query, hidden_df) filtered_df["Model"] = filtered_df.apply(add_model_hyperlinks, axis=1) df = select_columns(filtered_df, columns) return df def search_table(df: pd.DataFrame, query: str) -> pd.DataFrame: print(query) return df[(df[AutoEvalColumn.eval_name.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.eval_name.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 filter_queries(query: str, filtered_df: pd.DataFrame) -> pd.DataFrame: final_df = [] if query != "" and query is not None: 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.eval_name.name] ) return filtered_df def filter_models( df: pd.DataFrame, type_query: list, size_query: list, precision_query: list, show_deleted: bool ) -> pd.DataFrame: # Show all models #if show_deleted: # filtered_df = df #else: # Show only still on the hub models # filtered_df = df[df[AutoEvalColumn.still_on_hub.name] == True] filtered_df = df #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 def validate_upload(input): try: with open(input, mode="r") as f: data = json.load(f) #raise gr.Error("Cannot divide by zero!") except: raise gr.Error("Cannot parse file") def generate_captcha(width=300, height=220, length=4): text = ''.join(random.choices(string.ascii_uppercase + string.digits, k=length)) captcha_obj = ImageCaptcha(width, height) data = captcha_obj.generate(text) image = Image.open(data) return image, text def validate_captcha(input, text, img): img, new_text = generate_captcha() if input.lower() == text.lower(): return True, styled_message("Correct! You can procede with your submission."), new_text, img, "" return False, styled_error("Incorrect! Please retry with the new code."), new_text, img, "" demo = gr.Blocks(css=custom_css, theme=gr.themes.Default(primary_hue=gr.themes.colors.orange, secondary_hue=gr.themes.colors.orange)) 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("🏅 Leaderboard", elem_id="llm-benchmark-tab-table", id=0) as tb_board: 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 # Causes errors ], 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, ) gr.Markdown(TABLE_DESC, elem_classes="markdown-text") leaderboard_table = gr.Dataframe( value=leaderboard_df[ [c.name for c in fields(AutoEvalColumn) if c.never_hidden] + shown_columns.value ], headers=[c.name for c in fields(AutoEvalColumn) if c.never_hidden] + shown_columns.value, datatype=TYPES_LITE, elem_id="leaderboard-table", interactive=False, visible=True, wrap=False, ) # Dummy leaderboard for handling the case when the user uses backspace key hidden_leaderboard_table_for_search = gr.Dataframe( value=original_df[COLS], headers=COLS, datatype=TYPES, visible=False, ) search_bar.submit( update_table, [ hidden_leaderboard_table_for_search, shown_columns, search_bar, ], leaderboard_table, ) shown_columns.change( update_table, [ hidden_leaderboard_table_for_search, shown_columns, search_bar, ], leaderboard_table, queue=True, ) model_num = len(original_df) graph_df = original_df.drop(columns=[" ", "Precision", "Model URL"]).set_index("Model").T graph_ax = graph_df.plot( kind="barh", title="Graphical performance comparison", xlabel="Accuracy [%]", ylabel="Model", width=0.9, figsize=(15, 7 + 2*model_num), ) graph_ax.invert_yaxis() for container in graph_ax.containers: graph_ax.bar_label(container, fontsize=8, fmt="%.1f") graph_ax.legend(loc='center left', bbox_to_anchor=(1.01, 0.95)) plt.tight_layout(rect=[0, 0, 0.95, 1]) plot = gr.Plot(graph_ax.get_figure(), label="Graphical performance comparison") 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.Row(): gr.Markdown("# ✉️✨ Submit your model here!", elem_classes="markdown-text") with gr.Row(): with gr.Column(): with gr.Group(): model_name_textbox = gr.Textbox(label="Model name", info="Please avoid using the slash (/) character") #precision = gr.Radio(["bfloat16", "float16", "4bit"], label="Precision", info="What precision are you using for inference?") precision = gr.Dropdown( choices=[i.value.name for i in Precision if i != Precision.Unknown], label="Precision", multiselect=False, value="other", interactive=True, info="What weight precision were you using during the evaluation?" ) hf_model_id = gr.Textbox(label="Model link (Optional)", info="URL to the model's Hugging Face repository, or it's official website") contact_email = gr.Textbox(label="Your E-Mail") file_input = gr.File(file_count="single", interactive=True, label="Upload json file with evaluation results", file_types=['.json', '.jsonl']) file_input.upload(validate_upload, file_input) #upload_button = gr.UploadButton("Upload json", file_types=['.json']) #upload_button.upload(validate_upload, upload_button, file_input) with gr.Group(): captcha_correct = gr.State(False) text = gr.State("") image, text.value = generate_captcha() captcha_img = gr.Image( image, label="Prove your humanity", interactive=False, show_download_button=False, show_fullscreen_button=False, show_share_button=False, ) captcha_input = gr.Textbox(placeholder="Enter the text in the image above", show_label=False, container=False) check_button = gr.Button("Validate", interactive=True) captcha_result = gr.Markdown() check_button.click( fn = validate_captcha, inputs = [captcha_input, text, captcha_img], outputs = [captcha_correct, captcha_result, text, captcha_img, captcha_input], ) submit_button = gr.Button("Submit Eval", interactive=True) submission_result = gr.Markdown() submit_button.click( fn = add_new_eval, inputs = [ model_name_textbox, file_input, precision, hf_model_id, contact_email, captcha_correct, ], outputs = [submission_result, captcha_correct], ) with gr.Row(): with gr.Accordion("📙 Citation", open=False): citation_button = gr.Textbox( value=CITATION_BUTTON_TEXT, label=CITATION_BUTTON_LABEL, lines=14, elem_id="citation-button", show_copy_button=True, ) demo.load( fn=generate_captcha, outputs=[captcha_img, text] ) scheduler = BackgroundScheduler() scheduler.add_job(restart_space, "interval", seconds=43200) scheduler.start() demo.queue(default_concurrency_limit=40).launch(server_name="0.0.0.0")