app.py

import gradio as gr 
from datasets import load_dataset
import torch.nn.functional as F
import numpy as np
import torch
from transformers import AutoTokenizer, AutoModel
import pickle

with open('book_embeddings.pkl', 'rb') as file:
    book_embeddings = pickle.load(file)

model_checkpoint = 'intfloat/multilingual-e5-large'

tokenizer = AutoTokenizer.from_pretrained(model_checkpoint)
model = AutoModel.from_pretrained(model_checkpoint)

books_data = load_dataset('vojtam/czech_books_descriptions', split="train+test")
books_data.set_format('pandas')

def average_pool(last_hidden_states, attention_mask):
    last_hidden = last_hidden_states.masked_fill(~attention_mask[..., None].bool(), 0.0)
    return last_hidden.sum(dim=1) / attention_mask.sum(dim=1)[..., None]

def create_embeddings(tokenizer, model, input_texts, batch_size=32):
    embeddings_list = []
    
    for i in range(0, len(input_texts), batch_size):
        batch_texts = input_texts[i:i + batch_size]
        
        batch_dict = tokenizer(batch_texts, max_length=512, padding=True, truncation=True, return_tensors='pt')
        
        # Get embeddings for batch
        with torch.no_grad(): 
            outputs = model(**batch_dict)
            batch_embeddings = average_pool(outputs.last_hidden_state, batch_dict['attention_mask'])
            batch_embeddings = F.normalize(batch_embeddings, p=2, dim=1)
        
        embeddings_list.append(batch_embeddings)

        if (i + batch_size) % (batch_size * 10) == 0:
            print(f"Processed {i + batch_size}/{len(input_texts)} texts")
            
    return torch.cat(embeddings_list, dim=0)

def find_similar_books(query: str, n = 5):
    input_query = "query: " + query
    query_embedding = create_embeddings(tokenizer, model, input_query)
    scores = ((query_embedding @ book_embeddings.T) * 100).detach().numpy()[0]
    top_indices = np.argsort(scores)[-n:][::-1]

    return  books_data[top_indices]



css = """
.full-height-gallery {
    height: calc(100vh - 250px);
    overflow-y: auto;
}
#submit-btn {
    background-color: #ff5b00;
    color: #ffffff;
}
"""

with gr.Blocks(css=css) as intf:
    with gr.Row():
        text_input = gr.Textbox(label="Popis knihy", info = "Zadejte popis knihy, kterou byste si chtěli přečíst a aplikace najde nejpodobněší knihy dle vašeho popisu", placeholder='Zadejte popis, například "drama z prostředí nemocnice"')
        n_books = gr.Number(value = 5, label = "Počet knih", info="Počet nejpodobnějších knih, které si přejete zobrazit", minimum = 1, step = 1)
    with gr.Row():
        submit_btn = gr.Button("Vyhledat knihy", elem_id="submit-btn")
        clear_btn = gr.Button("Smazat")
    with gr.Row():
        dataframe = gr.Dataframe(label="Podobné knihy", show_label=False, elem_classes = ["full-height-gallery"])
    
    submit_btn.click(fn=find_similar_books, inputs=[text_input, n_books], outputs=dataframe)
    clear_btn.click(fn=lambda: [None, []], inputs=None, outputs=[text_input, dataframe])

intf.launch(share=True)