import uuid import gradio as gr from io_utils import get_logs_file, read_scanners, write_scanners from text_classification_ui_helpers import ( get_related_datasets_from_leaderboard, align_columns_and_show_prediction, check_dataset, deselect_run_inference, precheck_model_ds_enable_example_btn, select_run_mode, try_submit, write_column_mapping_to_config, ) from wordings import CONFIRM_MAPPING_DETAILS_MD, INTRODUCTION_MD MAX_LABELS = 40 MAX_FEATURES = 20 EXAMPLE_MODEL_ID = "cardiffnlp/twitter-roberta-base-sentiment-latest" EXAMPLE_DATA_ID = "tweet_eval" CONFIG_PATH = "./config.yaml" def get_demo(): with gr.Row(): gr.Markdown(INTRODUCTION_MD) uid_label = gr.Textbox( label="Evaluation ID:", value=uuid.uuid4, visible=False, interactive=False ) with gr.Row(): model_id_input = gr.Textbox( label="Hugging Face model id", placeholder=EXAMPLE_MODEL_ID + " (press enter to confirm)", ) with gr.Column(): no_dataset_checkbox = gr.Checkbox(label="Recommend a dataset", value=False, visible=True) dataset_id_input = gr.Textbox( label="Hugging Face Dataset id", placeholder=EXAMPLE_DATA_ID + " (press enter to confirm)", ) with gr.Row(): dataset_config_input = gr.Dropdown(label="Dataset Config", visible=False, allow_custom_value=True) dataset_split_input = gr.Dropdown(label="Dataset Split", visible=False, allow_custom_value=True) with gr.Row(): first_line_ds = gr.DataFrame(label="Dataset preview", visible=False) with gr.Row(): loading_status = gr.HTML(visible=True) with gr.Row(): example_btn = gr.Button( "Validate model & dataset", visible=True, variant="primary", interactive=False, ) with gr.Row(): example_input = gr.HTML(visible=False) with gr.Row(): example_prediction = gr.Label(label="Model Prediction Sample", visible=False) with gr.Row(): with gr.Accordion( label="Label and Feature Mapping", visible=False, open=False ) as column_mapping_accordion: with gr.Row(): gr.Markdown(CONFIRM_MAPPING_DETAILS_MD) column_mappings = [] with gr.Row(): with gr.Column(): gr.Markdown("# Label Mapping") for _ in range(MAX_LABELS): column_mappings.append(gr.Dropdown(visible=False)) with gr.Column(): gr.Markdown("# Feature Mapping") for _ in range(MAX_LABELS, MAX_LABELS + MAX_FEATURES): column_mappings.append(gr.Dropdown(visible=False)) with gr.Accordion(label="Model Wrap Advance Config (optional)", open=False): run_local = gr.Checkbox(value=True, label="Run in this Space") run_inference = gr.Checkbox(value=False, label="Run with Inference API") inference_token = gr.Textbox( value="", label="HF Token for Inference API", visible=False, interactive=True, ) with gr.Accordion(label="Scanner Advance Config (optional)", open=False): scanners = gr.CheckboxGroup(label="Scan Settings", visible=True) @gr.on(triggers=[uid_label.change], inputs=[uid_label], outputs=[scanners]) def get_scanners(uid): selected = read_scanners(uid) # currently we remove data_leakage from the default scanners # Reason: data_leakage barely raises any issues and takes too many requests # when using inference API, causing rate limit error scan_config = selected + ["data_leakage"] return gr.update( choices=scan_config, value=selected, label="Scan Settings", visible=True ) with gr.Row(): run_btn = gr.Button( "Get Evaluation Result", variant="primary", interactive=False, size="lg", ) with gr.Row(): logs = gr.Textbox( value=get_logs_file, label="Giskard Bot Evaluation Log:", visible=False, every=0.5, ) dataset_id_input.change( check_dataset, inputs=[dataset_id_input], outputs=[dataset_config_input, dataset_split_input, first_line_ds, loading_status], ) dataset_config_input.change( check_dataset, inputs=[dataset_id_input, dataset_config_input], outputs=[dataset_config_input, dataset_split_input, first_line_ds, loading_status], ) dataset_split_input.change( check_dataset, inputs=[dataset_id_input, dataset_config_input, dataset_split_input], outputs=[dataset_config_input, dataset_split_input, first_line_ds, loading_status], ) scanners.change(write_scanners, inputs=[scanners, uid_label]) run_inference.change( select_run_mode, inputs=[run_inference], outputs=[inference_token, run_local], ) run_local.change( deselect_run_inference, inputs=[run_local], outputs=[inference_token, run_inference], ) gr.on( triggers=[model_id_input.change, no_dataset_checkbox.change], fn=get_related_datasets_from_leaderboard, inputs=[model_id_input, no_dataset_checkbox], outputs=[dataset_id_input], ) gr.on( triggers=[label.change for label in column_mappings], fn=write_column_mapping_to_config, inputs=[ uid_label, *column_mappings, ], ) # label.change sometimes does not pass the changed value gr.on( triggers=[label.input for label in column_mappings], fn=write_column_mapping_to_config, inputs=[ uid_label, *column_mappings, ], ) gr.on( triggers=[ model_id_input.change, dataset_id_input.change, dataset_config_input.change, dataset_split_input.change, ], fn=precheck_model_ds_enable_example_btn, inputs=[ model_id_input, dataset_id_input, dataset_config_input, dataset_split_input, ], outputs=[example_btn, loading_status], ) gr.on( triggers=[ example_btn.click, ], fn=align_columns_and_show_prediction, inputs=[ model_id_input, dataset_id_input, dataset_config_input, dataset_split_input, uid_label, ], outputs=[ example_input, example_prediction, column_mapping_accordion, run_btn, loading_status, *column_mappings, ], ) gr.on( triggers=[ run_btn.click, ], fn=try_submit, inputs=[ model_id_input, dataset_id_input, dataset_config_input, dataset_split_input, run_local, run_inference, inference_token, uid_label, ], outputs=[run_btn, logs, uid_label], ) def enable_run_btn(): return gr.update(interactive=True) gr.on( triggers=[ run_inference.input, run_local.input, inference_token.input, scanners.input, ], fn=enable_run_btn, inputs=None, outputs=[run_btn], ) gr.on( triggers=[label.input for label in column_mappings], fn=enable_run_btn, inputs=None, # FIXME outputs=[run_btn], )