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_1, LLM_BENCHMARKS_TEXT_2, CROSS_EVALUATION_METRICS, NOTE_GENERATION_METRICS, # EVALUATION_EXAMPLE_IMG, # LLM_BENCHMARKS_TEXT_2, # ENTITY_DISTRIBUTION_IMG, # LLM_BENCHMARKS_TEXT_3, TITLE, LOGO, FIVE_PILLAR_DIAGRAM ) from src.display.css_html_js import custom_css # changes to be made here from src.display.utils import ( DATASET_BENCHMARK_COLS, OPEN_ENDED_BENCHMARK_COLS, MED_SAFETY_BENCHMARK_COLS, MEDICAL_SUMMARIZATION_BENCHMARK_COLS, ACI_BENCHMARK_COLS, SOAP_BENCHMARK_COLS, DATASET_COLS, OPEN_ENDED_COLS, MED_SAFETY_COLS, MEDICAL_SUMMARIZATION_COLS, ACI_COLS, SOAP_COLS, EVAL_COLS, EVAL_TYPES, NUMERIC_INTERVALS, TYPES, AutoEvalColumn, ModelType, ModelArch, PromptTemplateName, Precision, WeightType, fields, render_generation_templates ) 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, PLACEHOLDER_DATASET_WISE_NORMALIZATION_CONFIG 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() # Span based results # changes to be made here _, harness_datasets_original_df = get_leaderboard_df(EVAL_RESULTS_PATH, EVAL_REQUESTS_PATH, DATASET_COLS, DATASET_BENCHMARK_COLS, "accuracy", "datasets") harness_datasets_leaderboard_df = harness_datasets_original_df.copy() _, open_ended_original_df = get_leaderboard_df(EVAL_RESULTS_PATH, EVAL_REQUESTS_PATH, OPEN_ENDED_COLS, OPEN_ENDED_BENCHMARK_COLS, "score", "open_ended") open_ended_leaderboard_df = open_ended_original_df.copy() _, med_safety_original_df = get_leaderboard_df(EVAL_RESULTS_PATH, EVAL_REQUESTS_PATH, MED_SAFETY_COLS, MED_SAFETY_BENCHMARK_COLS, "score", "med_safety") med_safety_leaderboard_df = med_safety_original_df.copy() _, medical_summarization_original_df = get_leaderboard_df(EVAL_RESULTS_PATH, EVAL_REQUESTS_PATH, MEDICAL_SUMMARIZATION_COLS, MEDICAL_SUMMARIZATION_BENCHMARK_COLS, "score", "medical_summarization") medical_summarization_leaderboard_df = medical_summarization_original_df.copy() _, aci_original_df = get_leaderboard_df(EVAL_RESULTS_PATH, EVAL_REQUESTS_PATH, ACI_COLS, ACI_BENCHMARK_COLS, "score", "aci") aci_leaderboard_df = aci_original_df.copy() _, soap_original_df = get_leaderboard_df(EVAL_RESULTS_PATH, EVAL_REQUESTS_PATH, SOAP_COLS, SOAP_BENCHMARK_COLS, "score", "soap") soap_leaderboard_df = soap_original_df.copy() # breakpoint() # # Token based results # _, token_based_datasets_original_df = get_leaderboard_df(EVAL_RESULTS_PATH, EVAL_REQUESTS_PATH, DATASET_COLS, DATASET_BENCHMARK_COLS, "TokenBasedWithMacroAverage", "datasets") # token_based_datasets_leaderboard_df = token_based_datasets_original_df.copy() # _, token_based_types_original_df = get_leaderboard_df(EVAL_RESULTS_PATH, EVAL_REQUESTS_PATH, Clinical_TYPES_COLS, TYPES_BENCHMARK_COLS, "TokenBasedWithMacroAverage", "clinical_types") # token_based_types_leaderboard_df = token_based_types_original_df.copy() ( finished_eval_queue_df, running_eval_queue_df, pending_eval_queue_df, ) = get_evaluation_queue_df(EVAL_REQUESTS_PATH, EVAL_COLS) # breakpoint() def update_df(shown_columns, subset="datasets"): # changes to be made here if subset == "datasets": leaderboard_table_df = harness_datasets_leaderboard_df.copy() hidden_leader_board_df = harness_datasets_original_df elif subset == "open_ended": leaderboard_table_df = open_ended_leaderboard_df.copy() hidden_leader_board_df = open_ended_original_df elif subset == "med_safety": leaderboard_table_df = med_safety_leaderboard_df.copy() hidden_leader_board_df = med_safety_original_df elif subset == "medical_summarization": leaderboard_table_df = medical_summarization_leaderboard_df.copy() hidden_leader_board_df = medical_summarization_original_df elif subset == "aci": leaderboard_table_df = aci_leaderboard_df.copy() hidden_leader_board_df = aci_original_df elif subset == "soap": leaderboard_table_df = soap_leaderboard_df.copy() hidden_leader_board_df = soap_original_df # else: # match evaluation_metric: # case "Span Based": # leaderboard_table_df = span_based_types_leaderboard_df.copy() # hidden_leader_board_df = span_based_types_original_df # case "Token Based": # leaderboard_table_df = token_based_types_leaderboard_df.copy() # hidden_leader_board_df = token_based_types_original_df # case _: # pass value_cols = [c.name for c in fields(AutoEvalColumn) if c.never_hidden] + shown_columns # breakpoint() return leaderboard_table_df[value_cols], hidden_leader_board_df # Searching and filtering def update_table( hidden_df: pd.DataFrame, columns: list, query: str = "", type_query: list = None, domain_specific_query: list = None, size_query: list = None, precision_query: str = None, show_deleted: bool = False, ): filtered_df = filter_models(hidden_df, type_query, domain_specific_query, size_query, precision_query, show_deleted) filtered_df = filter_queries(query, filtered_df) df = select_columns(filtered_df, columns, list(hidden_df.columns)) return df def search_table(df: pd.DataFrame, query: str) -> pd.DataFrame: return df[(df[AutoEvalColumn.model.name].str.contains(query, case=False))] def select_columns(df: pd.DataFrame, columns: list, cols: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]] return filtered_df 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) 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, domain_specific_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 if type_query is not None: type_name = [t.split(" ")[1] for t in type_query] filtered_df = filtered_df.loc[df[AutoEvalColumn.model_type.name].isin(type_name)] if domain_specific_query is not None: domain_specifics = [] if "🏥 Clinical models" in domain_specific_query: domain_specifics.append(True) if "Generic models" in domain_specific_query: domain_specifics.append(False) filtered_df = filtered_df.loc[df[AutoEvalColumn.is_domain_specific.name].isin(domain_specifics)] # if architecture_query is not None: # arch_types = [t for t in architecture_query] # filtered_df = filtered_df.loc[df[AutoEvalColumn.architecture.name].isin(arch_types)] # # filtered_df = filtered_df.loc[df[AutoEvalColumn.architecture.name].isin(architecture_query + ["None"])] if precision_query is not None: if AutoEvalColumn.precision.name in df.columns: filtered_df = filtered_df.loc[df[AutoEvalColumn.precision.name].isin(precision_query + ["None"])] if size_query is not 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 demo = gr.Blocks(css=custom_css) with demo: print("hello") gr.HTML(TITLE) gr.HTML(LOGO) gr.Markdown(INTRODUCTION_TEXT, elem_classes="markdown-text") with gr.Tabs(elem_classes="tab-buttons") as tabs: with gr.TabItem("🏅 Closed Ended Evaluation", 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 (c.invariant or c.dataset_task_col)], value=[ c.name for c in fields(AutoEvalColumn) if c.displayed_by_default and not c.hidden and not c.never_hidden and (c.invariant or c.dataset_task_col) ], label="Select columns to show", elem_id="column-select", interactive=True, ) # with gr.Row(): # deleted_models_visibility = gr.Checkbox( # value=False, label="Show gated/private/deleted 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_architecture = gr.CheckboxGroup( # label="Architecture Types", # choices=[i.value.name for i in ModelArch], # value=[i.value.name for i in ModelArch], # interactive=True, # elem_id="filter-columns-architecture", # ) filter_domain_specific = gr.CheckboxGroup( label="Domain Specificity", choices=["🏥 Clinical models", "Generic models"], value=["🏥 Clinical models", "Generic models"], interactive=True, elem_id="filter-columns-type", ) 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", ) datasets_leaderboard_df, datasets_original_df = update_df(shown_columns.value, subset="datasets") leaderboard_table = gr.components.Dataframe( value=datasets_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, 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=datasets_original_df[DATASET_COLS], headers=DATASET_COLS, datatype=TYPES, visible=False, ) search_bar.submit( update_table, [ hidden_leaderboard_table_for_search, shown_columns, search_bar, filter_columns_type, filter_domain_specific, filter_columns_size # filter_columns_architecture ], leaderboard_table, ) for selector in [ shown_columns, filter_columns_type, filter_domain_specific, # filter_columns_architecture, filter_columns_size, # deleted_models_visibility, ]: selector.change( update_table, [ hidden_leaderboard_table_for_search, shown_columns, search_bar, filter_columns_type, filter_domain_specific, filter_columns_size # filter_columns_architecture, ], leaderboard_table, queue=True, ) with gr.TabItem("🏅 Open Ended Evaluation", elem_id="llm-benchmark-tab-table", id=1): 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 (c.invariant or c.open_ended_col)], value=[ c.name for c in fields(AutoEvalColumn) if c.displayed_by_default and not c.hidden and not c.never_hidden and (c.invariant or c.open_ended_col) ], label="Select columns to show", elem_id="column-select", interactive=True, ) # with gr.Row(): # deleted_models_visibility = gr.Checkbox( # value=False, label="Show gated/private/deleted 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_architecture = gr.CheckboxGroup( # label="Architecture Types", # choices=[i.value.name for i in ModelArch], # value=[i.value.name for i in ModelArch], # interactive=True, # elem_id="filter-columns-architecture", # ) filter_domain_specific = gr.CheckboxGroup( label="Domain Specificity", choices=["🏥 Clinical models", "Generic models"], value=["🏥 Clinical models", "Generic models"], interactive=True, elem_id="filter-columns-type", ) 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", ) datasets_leaderboard_df, datasets_original_df = update_df(shown_columns.value, subset="open_ended") leaderboard_table = gr.components.Dataframe( value=datasets_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, 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=datasets_original_df[OPEN_ENDED_COLS], headers=OPEN_ENDED_COLS, datatype=TYPES, visible=False, ) search_bar.submit( update_table, [ hidden_leaderboard_table_for_search, shown_columns, search_bar, filter_columns_type, filter_domain_specific, filter_columns_size # filter_columns_architecture ], leaderboard_table, ) for selector in [ shown_columns, filter_columns_type, filter_domain_specific, # filter_columns_architecture, filter_columns_size, # deleted_models_visibility, ]: selector.change( update_table, [ hidden_leaderboard_table_for_search, shown_columns, search_bar, filter_columns_type, filter_domain_specific, filter_columns_size # filter_columns_architecture, ], leaderboard_table, queue=True, ) with gr.Accordion("💬 Generation templates", open=False): with gr.Accordion("Response generation", open=False): system_prompt, user_prompt = render_generation_templates(task="open_ended", generation_type="response_generation") with gr.Accordion("Scoring Rubric", open=False): system_prompt, user_prompt = render_generation_templates(task="open_ended", generation_type="scoring_rubric") with gr.TabItem("🏅 Med Safety", elem_id="llm-benchmark-tab-table", id=2): 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 (c.invariant or c.med_safety_col)], value=[ c.name for c in fields(AutoEvalColumn) if c.displayed_by_default and not c.hidden and not c.never_hidden and (c.invariant or c.med_safety_col) ], label="Select columns to show", elem_id="column-select", interactive=True, ) # with gr.Row(): # deleted_models_visibility = gr.Checkbox( # value=False, label="Show gated/private/deleted 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_architecture = gr.CheckboxGroup( # label="Architecture Types", # choices=[i.value.name for i in ModelArch], # value=[i.value.name for i in ModelArch], # interactive=True, # elem_id="filter-columns-architecture", # ) filter_domain_specific = gr.CheckboxGroup( label="Domain Specificity", choices=["🏥 Clinical models", "Generic models"], value=["🏥 Clinical models", "Generic models"], interactive=True, elem_id="filter-columns-type", ) 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", ) datasets_leaderboard_df, datasets_original_df = update_df(shown_columns.value, subset="med_safety") leaderboard_table = gr.components.Dataframe( value=datasets_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, 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=datasets_original_df[MED_SAFETY_COLS], headers=MED_SAFETY_COLS, datatype=TYPES, visible=False, ) search_bar.submit( update_table, [ hidden_leaderboard_table_for_search, shown_columns, search_bar, filter_columns_type, filter_domain_specific, filter_columns_size # filter_columns_architecture ], leaderboard_table, ) for selector in [ shown_columns, filter_columns_type, filter_domain_specific, filter_columns_size, # deleted_models_visibility, ]: selector.change( update_table, [ hidden_leaderboard_table_for_search, shown_columns, search_bar, filter_columns_type, filter_domain_specific, filter_columns_size ], leaderboard_table, queue=True, ) with gr.Accordion("💬 Generation templates", open=False): with gr.Accordion("Response generation", open=False): system_prompt, user_prompt = render_generation_templates(task="med_safety", generation_type="response_generation") with gr.Accordion("Scoring Rubric", open=False): system_prompt, user_prompt = render_generation_templates(task="med_safety", generation_type="scoring_rubric") with gr.TabItem("🏅 Medical Summarization", elem_id="llm-benchmark-tab-table", id=3): gr.Markdown(CROSS_EVALUATION_METRICS, elem_classes="markdown-text") 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 (c.invariant or c.medical_summarization_col)], value=[ c.name for c in fields(AutoEvalColumn) if c.displayed_by_default and not c.hidden and not c.never_hidden and (c.invariant or c.medical_summarization_col) ], label="Select columns to show", elem_id="column-select", interactive=True, ) # with gr.Row(): # deleted_models_visibility = gr.Checkbox( # value=False, label="Show gated/private/deleted 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_architecture = gr.CheckboxGroup( # label="Architecture Types", # choices=[i.value.name for i in ModelArch], # value=[i.value.name for i in ModelArch], # interactive=True, # elem_id="filter-columns-architecture", # ) filter_domain_specific = gr.CheckboxGroup( label="Domain Specificity", choices=["🏥 Clinical models", "Generic models"], value=["🏥 Clinical models", "Generic models"], interactive=True, elem_id="filter-columns-type", ) 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", ) datasets_leaderboard_df, datasets_original_df = update_df(shown_columns.value, subset="medical_summarization") leaderboard_table = gr.components.Dataframe( value=datasets_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, 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=datasets_original_df[MEDICAL_SUMMARIZATION_COLS], headers=MEDICAL_SUMMARIZATION_COLS, datatype=TYPES, visible=False, ) search_bar.submit( update_table, [ hidden_leaderboard_table_for_search, shown_columns, search_bar, filter_columns_type, filter_domain_specific, filter_columns_size # filter_columns_architecture ], leaderboard_table, ) for selector in [ shown_columns, filter_columns_type, filter_domain_specific, filter_columns_size, # deleted_models_visibility, ]: selector.change( update_table, [ hidden_leaderboard_table_for_search, shown_columns, search_bar, filter_columns_type, filter_domain_specific, filter_columns_size ], leaderboard_table, queue=True, ) with gr.Accordion("💬 Generation templates", open=False): with gr.Accordion("Response generation", open=False): system_prompt, user_prompt = render_generation_templates(task="medical_summarization", generation_type="response_generation") with gr.Accordion("Question generation", open=False): system_prompt, user_prompt = render_generation_templates(task="ce", generation_type="question_generation") with gr.Accordion("Cross Examination", open=False): system_prompt, user_prompt = render_generation_templates(task="ce", generation_type="cross_examination") with gr.TabItem("🏅 Note generation", elem_id="llm-benchmark-tab-table", id=4): gr.Markdown(NOTE_GENERATION_METRICS, elem_classes="markdown-text") with gr.Tabs(elem_classes="tab-buttons2") as tabs: with gr.TabItem("ACI Bench", elem_id="llm-benchmark-tab-table2", 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 (c.invariant or c.aci_col)], value=[ c.name for c in fields(AutoEvalColumn) if c.displayed_by_default and not c.hidden and not c.never_hidden and (c.invariant or c.aci_col) ], label="Select columns to show", elem_id="column-select", interactive=True, ) # with gr.Row(): # deleted_models_visibility = gr.Checkbox( # value=False, label="Show gated/private/deleted 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_architecture = gr.CheckboxGroup( # label="Architecture Types", # choices=[i.value.name for i in ModelArch], # value=[i.value.name for i in ModelArch], # interactive=True, # elem_id="filter-columns-architecture", # ) filter_domain_specific = gr.CheckboxGroup( label="Domain Specificity", choices=["🏥 Clinical models", "Generic models"], value=["🏥 Clinical models", "Generic models"], interactive=True, elem_id="filter-columns-type", ) 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", ) datasets_leaderboard_df, datasets_original_df = update_df(shown_columns.value, subset="aci") leaderboard_table = gr.components.Dataframe( value=datasets_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, 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=datasets_original_df[ACI_COLS], headers=ACI_COLS, datatype=TYPES, visible=False, ) search_bar.submit( update_table, [ hidden_leaderboard_table_for_search, shown_columns, search_bar, filter_columns_type, filter_domain_specific, filter_columns_size # filter_columns_architecture ], leaderboard_table, ) for selector in [ shown_columns, filter_columns_type, filter_domain_specific, filter_columns_size, # deleted_models_visibility, ]: selector.change( update_table, [ hidden_leaderboard_table_for_search, shown_columns, search_bar, filter_columns_type, filter_domain_specific, filter_columns_size ], leaderboard_table, queue=True, ) with gr.TabItem("SOAP Notes", elem_id="llm-benchmark-tab-table2", id=1): 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 (c.invariant or c.soap_col)], value=[ c.name for c in fields(AutoEvalColumn) if c.displayed_by_default and not c.hidden and not c.never_hidden and (c.invariant or c.soap_col) ], label="Select columns to show", elem_id="column-select", interactive=True, ) # with gr.Row(): # deleted_models_visibility = gr.Checkbox( # value=False, label="Show gated/private/deleted 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_architecture = gr.CheckboxGroup( # label="Architecture Types", # choices=[i.value.name for i in ModelArch], # value=[i.value.name for i in ModelArch], # interactive=True, # elem_id="filter-columns-architecture", # ) filter_domain_specific = gr.CheckboxGroup( label="Domain Specificity", choices=["🏥 Clinical models", "Generic models"], value=["🏥 Clinical models", "Generic models"], interactive=True, elem_id="filter-columns-type", ) 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", ) datasets_leaderboard_df, datasets_original_df = update_df(shown_columns.value, subset="soap") leaderboard_table = gr.components.Dataframe( value=datasets_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, 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=datasets_original_df[SOAP_COLS], headers=SOAP_COLS, datatype=TYPES, visible=False, ) search_bar.submit( update_table, [ hidden_leaderboard_table_for_search, shown_columns, search_bar, filter_columns_type, filter_domain_specific, filter_columns_size # filter_columns_architecture ], leaderboard_table, ) for selector in [ shown_columns, filter_columns_type, filter_domain_specific, filter_columns_size, # deleted_models_visibility, ]: selector.change( update_table, [ hidden_leaderboard_table_for_search, shown_columns, search_bar, filter_columns_type, filter_domain_specific, filter_columns_size ], leaderboard_table, queue=True, ) with gr.Accordion("💬 Generation templates", open=False): with gr.Accordion("ACI-Bench Response generation", open=False): system_prompt, user_prompt = render_generation_templates(task="aci", generation_type="response_generation") with gr.Accordion("SOAP Notes Response generation", open=False): system_prompt, user_prompt = render_generation_templates(task="soap", generation_type="response_generation") with gr.Accordion("Question generation", open=False): system_prompt, user_prompt = render_generation_templates(task="ce", generation_type="question_generation") with gr.Accordion("Cross Examination", open=False): system_prompt, user_prompt = render_generation_templates(task="ce", generation_type="cross_examination") with gr.TabItem("📝 About", elem_id="llm-benchmark-tab-table", id=5): gr.Markdown(LLM_BENCHMARKS_TEXT_1, elem_classes="markdown-text") gr.HTML(FIVE_PILLAR_DIAGRAM) gr.Markdown(LLM_BENCHMARKS_TEXT_2, elem_classes="markdown-text") # gr.HTML(EVALUATION_EXAMPLE_IMG, elem_classes="logo") # gr.Markdown(LLM_BENCHMARKS_TEXT_2, elem_classes="markdown-text") # gr.HTML(ENTITY_DISTRIBUTION_IMG, elem_classes="logo") # gr.Markdown(LLM_BENCHMARKS_TEXT_3, elem_classes="markdown-text") with gr.TabItem("🚀 Submit here! ", elem_id="llm-benchmark-tab-table", id=6): 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") 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="auto", interactive=True, ) weight_type = gr.Dropdown( choices=[i.value.name for i in WeightType], label="Weights type", multiselect=False, value=WeightType.Original.value.name, interactive=False, ) base_model_name_textbox = gr.Textbox(label="Base model (for delta or adapter weights)", interactive=False) with gr.Row(): domain_specific_toggle = gr.Checkbox( label="Domain specific", value=False, info="Is your model medically oriented?", ) chat_template_toggle = gr.Checkbox( label="Use chat template", value=False, info="Is your model a chat model?", ) 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, model_type, domain_specific_toggle, chat_template_toggle, precision, weight_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=1800) scheduler.start() demo.queue(default_concurrency_limit=40).launch(allowed_paths=['./assets/'])