app.py

import tensorflow as tf
import tensorflow_hub as hub
from tensorflow_text import SentencepieceTokenizer
import gradio as gr
import math

model_url = "https://tfhub.dev/google/universal-sentence-encoder-multilingual-large/3"
model = hub.load(model_url)

def embed_text(text: str) -> dict:
    embeddings = model(text)
    return embeddings.numpy().tolist()[0]

embed_text_inter = gr.Interface(
    fn = embed_text,
    inputs = "text",
    outputs = gr.JSON(),
    title = "Universal Sentence Encoder 3 Large"
)

def distance(text_1: str, text_2: str) -> float:
    embeddings_1 = embed_text(text_1)
    embeddings_2 = embed_text(text_2)
    
    dist = 0
    for i in range(len(embeddings_1)):
        dist += (embeddings_1[i] - embeddings_2[i]) ** 2
    dist = math.sqrt(dist)
    return dist


distance_inter = gr.Interface(
    fn = distance,
    inputs = ["text", "text"],
    outputs = "number",
    title = "Universal Sentence Encoder 3 Large"
)


iface = gr.TabbedInterface(
    interface_list=[embed_text_inter, distance_inter],
    title="Universal Sentence Encoder 3 Large"
)

iface.launch()