app.py

import evaluate
import sys
from pathlib import Path
from evaluate.utils import infer_gradio_input_types, json_to_string_type, parse_readme, parse_gradio_data, parse_test_cases


def launch_gradio_widget(metric):
    """Launches `metric` widget with Gradio."""

    try:
        import gradio as gr
    except ImportError as error:
        logger.error("To create a metric widget with Gradio make sure gradio is installed.")
        raise error

    local_path = Path(sys.path[0])
    # if there are several input types, use first as default.
    if isinstance(metric.features, list):
        (feature_names, feature_types) = zip(*metric.features[0].items())
    else:
        (feature_names, feature_types) = zip(*metric.features.items())
    gradio_input_types = infer_gradio_input_types(feature_types)

    def compute(data):
        return metric.compute(**parse_gradio_data(data, gradio_input_types))

    header_html = '''<div style="max-width:800px; margin:auto; float:center; margin-top:0; margin-bottom:0; padding:0;">
            <img src="https://huggingface.co/spaces/xu1998hz/sescore/resolve/main/img/logo_sescore.png" style="margin:0; padding:0; margin-top:-10px; margin-bottom:-50px;">
        </div>
        <h2 style='margin-top: 5pt; padding-top:10pt;'>About <i>SEScore</i></h2>

        <p><b>SEScore</b> is a reference-based text-generation evaluation metric that requires no pre-human-annotated error data, 
        described in our paper <a href="https://arxiv.org/abs/2210.05035"><b>"Not All Errors are Equal: Learning Text Generation Metrics using 
        Stratified Error Synthesis"</b></a> from EMNLP 2022.</p>

        <p>Its effectiveness over prior methods like BLEU, BERTScore, BARTScore, PRISM, COMET and BLEURT has been demonstrated on a diverse set of language generation tasks, including
        translation, captioning, and web text generation. <a href="https://twitter.com/LChoshen/status/1580136005654700033">Readers have even described SEScore as "one unsupervised evaluation to rule them all"</a> 
        and we are very excited to share it with you!</p>

        <h2 style='margin-top: 10pt; padding-top:0;'>Try it yourself!</h2>
        <p>Provide sample (gold) reference text and (model output) predicted text below and see how SEScore rates them! It is most performant
        in a relative ranking setting, so in general <b>it will rank better predictions higher than worse ones.</b> Providing useful
        absolute numbers based on SEScore is an ongoing direction of investigation.</p>
    '''.replace('\n',' ')


    tail_markdown = parse_readme(local_path / "description.md")


    iface = gr.Interface(
        fn=compute,
        inputs=gr.inputs.Dataframe(
            headers=feature_names,
            col_count=len(feature_names),
            row_count=2,
            datatype=json_to_string_type(gradio_input_types),
        ),
        outputs=gr.outputs.Textbox(label=metric.name),
        description=header_html,
        #title=f"SEScore Metric Usage Example",
        article=tail_markdown,
        # TODO: load test cases and use them to populate examples
        # examples=[parse_test_cases(test_cases, feature_names, gradio_input_types)]
    )

    print(dir(iface))

    iface.launch()



module = evaluate.load("xu1998hz/sescore")
launch_gradio_widget(module)