app.py

import gradio as gr
import os
from langchain.retrievers import EnsembleRetriever
from utils import *
import requests
from pyvi import ViTokenizer, ViPosTagger
import time
from transformers import AutoTokenizer, AutoModelForQuestionAnswering
import torch

retriever = load_the_embedding_retrieve(is_ready=False, k=3)
bm25_retriever = load_the_bm25_retrieve(k=3)

ensemble_retriever = EnsembleRetriever(
    retrievers=[bm25_retriever, retriever], weights=[0.5, 0.5]
)

tokenizer = AutoTokenizer.from_pretrained("ShynBui/vie_qa", token=os.environ.get("HF_TOKEN"))
model = AutoModelForQuestionAnswering.from_pretrained("ShynBui/vie_qa", token=os.environ.get("HF_TOKEN"))

headers = {
    "Accept": "application/json",
    "Authorization": "Bearer " + os.environ.get("HF_TOKEN"),
    "Content-Type": "application/json"
}


def query(payload):
    response = requests.post(os.environ.get("API_URL"), headers=headers, json=payload)
    return response.json()


def greet(quote):
    sources = []
    answers = []
    scores = []
    ids = []

    docs = ensemble_retriever.get_relevant_documents(quote)

    for i in docs:
        context = ViTokenizer.tokenize(i.page_content)
        question = ViTokenizer.tokenize(quote)
        print("source:", i.metadata['source'])
        sources.append(i.metadata['source'])
        output = query({
            "inputs": {
                "question": question,
                "context": context[:256]
            },
        })
        while "error" in output:
            # print('fail')
            time.sleep(1)
            output = query({
                "inputs": {
                    "question": question,
                    "context": context[:256]
                },
            })

        answers.append(output['answer'])
    return answers


def greet2(quote):
    answers = []
    docs = ensemble_retriever.get_relevant_documents(quote)

    return docs
    
    for i in docs:
        context = ViTokenizer.tokenize(i.page_content)
        question = ViTokenizer.tokenize(quote)

        inputs = tokenizer(question, context, return_tensors="pt")

        outputs = model(**inputs)

        start_index = torch.argmax(outputs.start_logits)
        end_index = torch.argmax(outputs.end_logits) + 1

        answer = tokenizer.convert_tokens_to_string(
            tokenizer.convert_ids_to_tokens(inputs["input_ids"][0][start_index:end_index]))

        answers.append(answer)

    return answers


if __name__ == "__main__":
    quote = "Địa chỉ nhà trường?"

    iface = gr.Interface(fn=greet2, inputs="text", outputs="text")
    iface.launch()