app.py

import time
import numpy as np
import pandas as pd
import torch
import faiss
from sklearn.preprocessing import normalize
from transformers import AutoTokenizer, AutoModelForQuestionAnswering
from sentence_transformers import SentenceTransformer
import pickle
import gradio as gr

print(torch.cuda.is_available())

__all__ = [
    "mdeberta",
    "wangchanberta-hyp",  # Best model
]

predict_method = [
    "faiss",
    "faissWithModel",
    "cosineWithModel",
    "semanticSearchWithModel",
]

DEFAULT_MODEL = 'wangchanberta-hyp'
DEFAULT_SENTENCE_EMBEDDING_MODEL = 'intfloat/multilingual-e5-base'

MODEL_DICT = {
    'wangchanberta': 'powerpuf-bot/wangchanberta-th-wiki-qa_ref-params',
    'wangchanberta-hyp': 'powerpuf-bot/wangchanberta-th-wiki-qa_hyp-params',
    'mdeberta': 'powerpuf-bot/mdeberta-v3-th-wiki-qa_ref-params',
    'mdeberta-hyp': 'powerpuf-bot/mdeberta-v3-th-wiki-qa_hyp-params',
}

DATA_PATH = 'models/dataset.xlsx'
EMBEDDINGS_PATH = 'models/embeddings.pkl'

class ChatbotModel:
    def __init__(self, model=DEFAULT_MODEL):
        self._chatbot = Chatbot()
        self._chatbot.load_data()
        self._chatbot.load_model(model)
        self._chatbot.load_embedding_model(DEFAULT_SENTENCE_EMBEDDING_MODEL)
        self._chatbot.set_vectors()
        self._chatbot.set_index()

    def chat(self, question):
        return self._chatbot.answer_question(question)

class Chatbot:
    def __init__(self):
        # Initialize variables
        self.df = None
        self.test_df = None
        self.model = None
        self.model_name = None
        self.tokenizer = None
        self.embedding_model = None
        self.vectors = None
        self.index = None
        self.k = 1  # top k most similar

    def load_data(self, path: str = DATA_PATH):
        self.df = pd.read_excel(path, sheet_name='Default')
        self.df['Context'] = pd.read_excel(path, sheet_name='mdeberta')['Context']

    def load_model(self, model_name: str = DEFAULT_MODEL):
        self.model = AutoModelForQuestionAnswering.from_pretrained(MODEL_DICT[model_name])
        self.tokenizer = AutoTokenizer.from_pretrained(MODEL_DICT[model_name])
        self.model_name = model_name

    def load_embedding_model(self, model_name: str = DEFAULT_SENTENCE_EMBEDDING_MODEL):
        if torch.cuda.is_available():
            self.embedding_model = SentenceTransformer(model_name, device='cuda')
        else:
            self.embedding_model = SentenceTransformer(model_name)

    def set_vectors(self):
        self.vectors = self.prepare_sentences_vector(self.load_embeddings(EMBEDDINGS_PATH))

    def set_index(self):
        if torch.cuda.is_available():
            res = faiss.StandardGpuResources()
            self.index = faiss.IndexFlatL2(self.vectors.shape[1])
            gpu_index_flat = faiss.index_cpu_to_gpu(res, 0, self.index)
            gpu_index_flat.add(self.vectors)
            self.index = gpu_index_flat
        else:
            self.index = faiss.IndexFlatL2(self.vectors.shape[1])
            self.index.add(self.vectors)

    def get_embeddings(self, text_list):
        return self.embedding_model.encode(text_list)

    def prepare_sentences_vector(self, encoded_list):
        encoded_list = [i.reshape(1, -1) for i in encoded_list]
        encoded_list = np.vstack(encoded_list).astype('float32')
        encoded_list = normalize(encoded_list)
        return encoded_list

    def store_embeddings(self, embeddings):
        with open('models/embeddings.pkl', "wb") as fOut:
            pickle.dump({'sentences': self.df['Question'], 'embeddings': embeddings}, fOut, protocol=pickle.HIGHEST_PROTOCOL)

    def load_embeddings(self, file_path):
        with open(file_path, "rb") as fIn:
            stored_data = pickle.load(fIn)
            stored_sentences = stored_data['sentences']
            stored_embeddings = stored_data['embeddings']
        return stored_embeddings

    def model_pipeline(self, question, similar_context):
        inputs = self.tokenizer(question, similar_context, return_tensors="pt")
        with torch.no_grad():
            outputs = self.model(**inputs)
        answer_start_index = outputs.start_logits.argmax()
        answer_end_index = outputs.end_logits.argmax()
        predict_answer_tokens = inputs.input_ids[0, answer_start_index: answer_end_index + 1]
        Answer = self.tokenizer.decode(predict_answer_tokens)
        return Answer

    def faiss_search(self, question_vector):
        distances, indices = self.index.search(question_vector, self.k)
        similar_questions = [self.df['Question'][indices[0][i]] for i in range(self.k)]
        similar_contexts = [self.df['Context'][indices[0][i]] for i in range(self.k)]
        return similar_questions, similar_contexts, distances, indices

    def predict(self, message):
        message = message.strip()
        question_vector = self.get_embeddings(message)
        question_vector = self.prepare_sentences_vector([question_vector])
        similar_questions, similar_contexts, distances, indices = self.faiss_search(question_vector)
        context = similar_contexts[0]
        Answer = self.model_pipeline(similar_questions[0], context)
        return Answer

if __name__ == "__main__":
    bot = ChatbotModel()
    
    def chat_interface(question, history):
        response = bot._chatbot.predict(question)
        return response

    EXAMPLE = ["หลิน ไห่เฟิง มีชื่อเรียกอีกชื่อว่าอะไร" , "ใครเป็นผู้ตั้งสภาเศรษฐกิจโลกขึ้นในปี พ.ศ. 2514 โดยทุกปีจะมีการประชุมที่ประเทศสวิตเซอร์แลนด์", "โปรดิวเซอร์ของอัลบั้มตลอดกาล ของวงคีรีบูนคือใคร", "สกุลเดิมของหม่อมครูนุ่ม นวรัตน ณ อยุธยา คืออะไร"]
    
    demo = gr.ChatInterface(fn=chat_interface, examples=EXAMPLE, title="CE66-04: Thai Question Answering System by using Deep Learning")
    demo.launch()