app.py

from statistics import mean
import random

import torch
from transformers import BertModel, BertTokenizerFast
import numpy as np
import torch.nn.functional as F
import gradio as gr

tokenizer = BertTokenizerFast.from_pretrained("setu4993/LaBSE")
model = BertModel.from_pretrained("setu4993/LaBSE")
model = model.eval()

def embed(text, tokenizer, model):
    inputs = tokenizer(text, return_tensors="pt", padding=True)
    with torch.no_grad():
        outputs = model(**inputs)
    return outputs.pooler_output

def similarity(embeddings_1, embeddings_2):
    normalized_embeddings_1 = F.normalize(embeddings_1, p=2)
    normalized_embeddings_2 = F.normalize(embeddings_2, p=2)
    return torch.matmul(
        normalized_embeddings_1, normalized_embeddings_2.transpose(0, 1)
    )

def semantic_sim(sentence1, sentence2):
    em1 = embed(sentence1, tokenizer, model)
    em2 = embed(sentence2, tokenizer, model)
    sim = int(float(similarity(em1, em2)*5))
    out = ""
    if sim == 5:
        out = "Equivalent"
    elif sim == 4:
        out = "Mostly equivalent, unimportant details differ"
    elif sim == 3:
        out = "Roughly equivalent, important details differ or are missing"
    elif sim == 2:
        out = "Not equivalent, but share some details"
    elif sim == 1:
        out = "Same general topic, but not equivalent"
    elif sim == 0:
        out = "Completely dissimilar"
    return out

iface = gr.Interface(fn=semantic_sim, inputs=["text", "text"], outputs=["text"]).launch()