app.py

import os
from email.utils import parseaddr

import gradio as gr
import pandas as pd

from datasets import load_dataset
from apscheduler.schedulers.background import BackgroundScheduler
from huggingface_hub import HfApi

# InfoStrings
from content import *

BALM_TOKEN = os.environ.get("BALM_TOKEN", None)
owner="clefourrier" # change to balm once possible

api = HfApi()

eval_results = {}
eval_dataframe = {}
for level in range(1, 4):
    eval_results[level] = load_dataset(f"{owner}/BALM_ResultsLevel{level}", token=BALM_TOKEN, split="dev")
    eval_dataframe[level] = pd.DataFrame(eval_results[level].remove_column("mail"))

def restart_space():
    api.restart_space(repo_id=f"{owner}/BALM_Leaderboard", token=BALM_TOKEN)


COLS = ["Model", "Organisation", "Reported accuracy ⬆️"]
TYPES = ["str", "str", "number",]

def add_new_eval(
    level_of_dev: str,
    model: str,
    score: float,
    organisation: str,
    mail: str,
):
    level = int(level_of_dev.split(" ")[-1])

    # Very basic email parsing
    _, parsed_mail = parseaddr(mail)
    if not "@" in parsed_mail:
        valid_mail = "Please provide a valid email adress."
        return f"<p style='color: orange; font-size: 20px; text-align: center;'>{valid_mail}</p>"

    print("Adding new eval")

    # Check if the combination model/org already exists and prints a warning message if yes
    if model.lower() in set(eval_results[level]["model"]) and organisation.lower() in set(eval_results[level]["organisation"]):
        duplicate_request_message = "This model has been already submitted."
        return f"<p style='color: orange; font-size: 20px; text-align: center;'>{duplicate_request_message}</p>"

    # Actual submission
    eval_entry = {
        "model": model,
        "score": score,
        "organisation": organisation,
        "mail": mail,
    }
    eval_results[level].add_item(eval_entry)

    success_message = f"Model {model} submitted by {organisation}."
    return f"<p style='color: green; font-size: 20px; text-align: center;'>{success_message}</p>"


def refresh():
    eval_results = {}
    eval_dataframe = {}
    for level in range(1, 4):
        eval_results[level] = load_dataset(f"{owner}/BALM_ResultsLevel{level}", token=BALM_TOKEN, split="dev")
        eval_dataframe[level] = pd.DataFrame(eval_results[level].remove_column("mail"))
    return eval_dataframe[1],  eval_dataframe[2],  eval_dataframe[3]


custom_css = """
#changelog-text {
    font-size: 16px !important;
}

#changelog-text h2 {
    font-size: 18px !important;
}

.markdown-text {
    font-size: 16px !important;
}

#citation-button span {
    font-size: 16px !important;
}

#citation-button textarea {
    font-size: 16px !important;
}

#citation-button > label > button {
    margin: 6px;
    transform: scale(1.3);
}
"""

demo = gr.Blocks(css=custom_css)
with demo:
    gr.HTML(TITLE)
    gr.Markdown(INTRODUCTION_TEXT, elem_classes="markdown-text")

    with gr.Row():
        with gr.Column():
            with gr.Accordion("📙 Citation", open=False):
                citation_button = gr.Textbox(
                    value=CITATION_BUTTON_TEXT,
                    label=CITATION_BUTTON_LABEL,
                    elem_id="citation-button",
                ).style(show_copy_button=True)
        with gr.Column():
            with gr.Accordion("✨ CHANGELOG", open=False):
                changelog = gr.Markdown(CHANGELOG_TEXT, elem_id="changelog-text")

    with gr.Tab("Results: Level 1"):
        with gr.Tab("Results on Dev Set"):
            leaderboard_table_1 = gr.components.Dataframe(
                value=eval_dataframe[1], headers=COLS, datatype=TYPES, max_rows=20
            )
        with gr.Tab("Results on Test Set"):
            gr.Textbox(value="The test set is currently private! Come back when performances on the dev set increased!")
    with gr.Tab("Results: Level 2"):
        with gr.Tab("Results on Dev Set"):
            leaderboard_table_2 = gr.components.Dataframe(
                value=eval_dataframe[2], headers=COLS, datatype=TYPES, max_rows=20
            )
        with gr.Tab("Results on Test Set"):
            gr.Textbox(value="The test set is currently private! Come back when performances on the dev set increased!")
    with gr.Tab("Results: Level 3"):
        with gr.Tab("Results on Dev Set"):
            leaderboard_table_3 = gr.components.Dataframe(
                value=eval_dataframe[3], headers=COLS, datatype=TYPES, max_rows=20
            )
        with gr.Tab("Results on Test Set"):
            gr.Textbox(value="The test set is currently private! Come back when performances on the dev set increased!")

    refresh_button = gr.Button("Refresh")
    refresh_button.click(
        refresh,
        inputs=[],
        outputs=[
            eval_dataframe[1],
            eval_dataframe[2],
            eval_dataframe[3],
        ],
    )

    with gr.Accordion("Submit a new model for evaluation"):
        #with gr.Row():
        with gr.Column():
            level_of_dev = gr.Radio(["Level 1", "Level 2", "Level 3"], value="Level 1", label="Dev set")
            model_name_textbox = gr.Textbox(label="Model name")
            score = gr.Textbox(label="Score")
            organisation = gr.Textbox(label="Organisation")
            mail = gr.Textbox(label="Contact email")

        submit_button = gr.Button("Submit Eval")
        submission_result = gr.Markdown()
        submit_button.click(
            add_new_eval,
            [
                level_of_dev,
                model_name_textbox,
                score,
                organisation,
                mail
            ],
            submission_result,
        )

scheduler = BackgroundScheduler()
scheduler.add_job(restart_space, "interval", seconds=3600)
scheduler.start()
demo.launch()