app.py

import argparse
import logging
import sentence_transformers
import datasets
import gradio as gr

logging.disable(logging.CRITICAL)


model = sentence_transformers.SentenceTransformer(
    "dangvantuan/sentence-camembert-large", device="cpu")


dataset = datasets.load_dataset("json", data_files=["./data/dataset.json"], split="train")
dataset.load_faiss_index("embeddings", "index.faiss")

def search(query, k):
    query_embedding = model.encode(query)
    _, retrieved_examples = dataset.get_nearest_examples(
        "embeddings",
        query_embedding,
        k=int(k),
    )
    results = []
    for text, start, end, title, url in zip(
        retrieved_examples["text"],
        retrieved_examples["start"],
        retrieved_examples["end"],
        retrieved_examples["title"],
        retrieved_examples["url"],
    ):
        start = start
        end = end
        result = {
            "title": title,
            "transcript": f"[{str(start)} ====> {str(end)}] {text}",
            "link": url,
        }
        results.append(result)
    return results

iface = gr.Interface(
    search,
    inputs=[
        gr.inputs.Textbox(label="Query"),
        gr.inputs.Number(label="K", default=3, min_value=1, max_value=10),
    ],
    outputs=[
        gr.outputs.Textbox(label="Result 1"),
        gr.outputs.Textbox(label="Result 2"),
        gr.outputs.Textbox(label="Result 3"),
    ],
    title="Camembert and Faiss-powered Search Engine",
    description="Search through a dataset using Camembert and Faiss",
    theme="light",
    layout="vertical",
)

iface.launch()