app.py

import gradio as gr
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
import requests

tokenizer = AutoTokenizer.from_pretrained("impresso-project/nel-hipe-multilingual")
model = AutoModelForSeq2SeqLM.from_pretrained(
    "impresso-project/nel-hipe-multilingual"
).eval()

print("Model loaded successfully!")


def get_wikipedia_title(qid, language="en"):
    url = f"https://www.wikidata.org/w/api.php"
    params = {
        "action": "wbgetentities",
        "format": "json",
        "ids": qid,
        "props": "sitelinks/urls",
        "sitefilter": f"{language}wiki",
    }

    response = requests.get(url, params=params)
    data = response.json()

    try:
        title = data["entities"][qid]["sitelinks"][f"{language}wiki"]["title"]
        url = data["entities"][qid]["sitelinks"][f"{language}wiki"]["url"]
        return title, url
    except KeyError:
        return "NIL", "None"


def disambiguate_sentence(sentence):
    entities = []
    # Generate model outputs for the sentence
    outputs = model.generate(
        **tokenizer([sentence], return_tensors="pt"),
        num_beams=5,
        num_return_sequences=5,
        max_new_tokens=30,
    )
    decoded = tokenizer.batch_decode(outputs, skip_special_tokens=True)
    qid = decoded[0].split()[-1]  # Assuming QID is the last token in the output

    # Get Wikipedia title and URL
    title, url = get_wikipedia_title(qid)

    entity_info = f"QID: {qid}, Title: {title}, URL: {url}"
    entities.append(entity_info)

    print(f"Entities: {entities}")
    return {"text": sentence, "entities": entities}


def nel_app_interface():
    input_sentence = gr.Textbox(
        lines=5,
        label="Input Sentence",
        placeholder="Enter your sentence here in the following format: //  << We are going to [START] Paris [END]. >>"
        " // This format ensures that the model knows which entities to disambiguate, more exactly the "
        "entity should be surrounded by `[START]` and `[END]`. // "
        "!Only one entity per sentence is supported at the moment!",
    )
    output_entities = gr.Textbox(label="Linked Entities")

    # Interface definition
    interface = gr.Interface(
        fn=disambiguate_sentence,
        inputs=input_sentence,
        outputs=output_entities,
        title="Entity Linking with impresso-project/nel-hipe-multilingual",
        description="Link entities using the `impresso-project/nel-hipe-multilingual` model under the hood!",
        examples=[
            [
                "Des chercheurs de l' [START] Université de Cambridge [END] ont développé une nouvelle technique de calcul "
                "quantique qui promet d'augmenter exponentiellement les vitesses de calcul.",
                "Le rapport complet sur ces découvertes a été publié dans la prestigieuse revue 'Nature Physics'. ([START] "
                "Reuters [END])",
                "In the [START] year 1789 [END], the Estates-General was convened in France.",
                "[START] King Louis XVI, ruler of France [END], called for the meeting.",
                "The event was held at the [START] Palace of Versailles [END], a symbol of French monarchy.",
                "At Versailles, Marie Antoinette, the Queen of France, was involved in discussions.",
                "Maximilien Robespierre, a leading member of the National Assembly, also participated.",
                "[START] Jean-Jacques Rousseau, the famous philosopher [END], was a significant figure in the debate.",
                "Another important participant was [START] Charles de Talleyrand, the Bishop of Autun [END].",
                "Meanwhile, across the Atlantic, [START] George Washington, the first President of the United States [END], "
                "was shaping policies.",
                "[START] Thomas Jefferson, the nation's Secretary of State [END], played a key role in drafting policies for "
                "the new American government.",
            ]
        ],
    )

    interface.launch()


if __name__ == "__main__":
    nel_app_interface()