app.py

from fastapi import FastAPI, UploadFile, File
from fastapi.responses import FileResponse
from datasets import load_dataset
from fastapi.middleware.cors import CORSMiddleware
import pdfplumber
import pytesseract

from models import Invoice

# Loading
import os
import zipfile
import shutil
from os import makedirs,getcwd
from os.path import join,exists,dirname
import torch
import json
from haystack_integrations.document_stores.qdrant import QdrantDocumentStore

app = FastAPI()

app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],
    allow_credentials=True,
    allow_methods=["*"],
    allow_headers=["*"],
)


NUM_PROC = os.cpu_count()
parent_path  = dirname(getcwd())

temp_path  = join(parent_path,'temp')
if not exists(temp_path ):
    makedirs(temp_path )

# Determine device based on GPU availability
device = "cuda" if torch.cuda.is_available() else "cpu"
print(f"Using device: {device}")

import logging

logging.basicConfig(format="%(levelname)s - %(name)s -  %(message)s", level=logging.WARNING)
logging.getLogger("haystack").setLevel(logging.INFO)

document_store = QdrantDocumentStore(
        path="database",
        recreate_index=True,
        use_sparse_embeddings=True,
        embedding_dim = 384
    )

def extract_zip(zip_path, target_folder):
    """
    Extracts all files from a ZIP archive and returns a list of their paths.

    Args:
        zip_path (str): Path to the ZIP file.
        target_folder (str): Folder where the files will be extracted.

    Returns:
        List[str]: List of extracted file paths.
    """
    extracted_files = []
    with zipfile.ZipFile(zip_path, 'r') as zip_ref:
        zip_ref.extractall(target_folder)
        for filename in zip_ref.namelist():
            extracted_files.append(os.path.join(target_folder, filename))
    return extracted_files


def extract_text_from_pdf(pdf_path):
    with pdfplumber.open(pdf_path) as pdf:
        text = ""
        for page in pdf.pages:
            text += page.extract_text()
    return text

def extract_ocr_text_from_pdf(pdf_path):
    from pdf2image import convert_from_path
    images = convert_from_path(pdf_path)

    text= ""
    
    for image in images:
        text += pytesseract.image_to_string(image,lang='vie')
        
    return text
    
@app.post("/uploadfile/")
async def create_upload_file(text_field: str, file: UploadFile = File(...), ocr:bool=False):
    # Imports
    import time
    from haystack import Document, Pipeline
    from haystack.components.writers import DocumentWriter    
    from haystack.components.preprocessors import DocumentSplitter, DocumentCleaner
    from haystack.components.joiners import DocumentJoiner

    from haystack_integrations.components.retrievers.qdrant import QdrantHybridRetriever
    from haystack.document_stores.types import DuplicatePolicy
    from haystack_integrations.components.embedders.fastembed import (
     FastembedTextEmbedder,
     FastembedDocumentEmbedder,
     FastembedSparseTextEmbedder,
     FastembedSparseDocumentEmbedder
    )
    
    start_time = time.time()
    
    file_savePath =  join(temp_path,file.filename)

    with open(file_savePath,'wb') as f:
        shutil.copyfileobj(file.file, f)
    
    documents=[]
    
    # Here you can save the file and do other operations as needed
    if '.json' in file_savePath:
        with open(file_savePath) as fd:
            for line in fd:
                obj = json.loads(line)
                document = Document(content=obj[text_field], meta=obj) 
                documents.append(document)

    elif '.zip' in file_savePath:
        extracted_files_list = extract_zip(file_savePath, temp_path)
        print("Extracted files:")
        for file_path in extracted_files_list:
            if '.pdf' in file_path:
                if ocr:
                    text = extract_ocr_text_from_pdf(file_path)
                else:                    
                    text = extract_text_from_pdf(file_path)
                obj = {text_field:text,file_path:file_path}
                document = Document(content=obj[text_field], meta=obj) 
                documents.append(document)
    else:
        raise NotImplementedError("This feature is not supported yet")

    # Indexing
    
    
    indexing = Pipeline()

    document_joiner = DocumentJoiner()


    document_cleaner = DocumentCleaner() 
    
    document_splitter = DocumentSplitter(split_by="passage", split_length=1000, split_overlap=0)

    indexing.add_component("document_joiner", document_joiner)
    indexing.add_component("document_cleaner", document_cleaner)
    indexing.add_component("document_splitter", document_splitter)
    indexing.add_component("sparse_doc_embedder", FastembedSparseDocumentEmbedder(model="Qdrant/bm42-all-minilm-l6-v2-attentions"))
    indexing.add_component("dense_doc_embedder", FastembedDocumentEmbedder(model="sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2"))
    indexing.add_component("writer", DocumentWriter(document_store=document_store, policy=DuplicatePolicy.OVERWRITE))

    
    indexing.connect("document_joiner", "document_cleaner")
    indexing.connect("document_cleaner", "document_splitter")
    indexing.connect("document_splitter", "sparse_doc_embedder")

    indexing.connect("sparse_doc_embedder", "dense_doc_embedder")
    indexing.connect("dense_doc_embedder", "writer")
    
    indexing.run({"document_joiner": {"documents": documents}})
    end_time = time.time()

    elapsed_time = end_time - start_time
    
    return {"filename": file.filename, "message": "Done", "execution_time": elapsed_time}

    
@app.get("/search")
def search(prompt: str):
    import time
    from haystack import Document, Pipeline
    from haystack_integrations.components.retrievers.qdrant import QdrantHybridRetriever
    from haystack_integrations.components.embedders.fastembed import (
     FastembedTextEmbedder,
     FastembedSparseTextEmbedder
    )
    from haystack.components.rankers import TransformersSimilarityRanker
    from haystack.components.joiners import DocumentJoiner
    from haystack.components.generators import OpenAIGenerator
    from haystack.utils import Secret    
    from haystack.components.builders import PromptBuilder
    
    start_time = time.time()
    
    # Querying

    
    template = """
        Với thông tin sau, hãy trả lời câu hỏi bằng tiếng Việt.
    
        Bối cảnh: {% for document in documents %}
        
        {{ document.content }}
        {% endfor %}
        
        Câu hỏi: {{ question }}
        
        Trả lời:
    """
    #######################################################
    """"
    Given the following information, answer the question.
    
    Context:
    {% for document in documents %}
        {{ document.content }}
    {% endfor %}
    
    Question: {{question}}
    Answer:
    """
    
    prompt_builder = PromptBuilder(template=template)
    generator = OpenAIGenerator(
        api_key=Secret.from_env_var("OCTOAI_TOKEN"),
        api_base_url="https://text.octoai.run/v1",
        model="meta-llama-3-70b-instruct",
        generation_kwargs = {"max_tokens": 512}
    )

    querying = Pipeline()
    querying.add_component("sparse_text_embedder", FastembedSparseTextEmbedder(model="Qdrant/bm42-all-minilm-l6-v2-attentions"))
    querying.add_component("dense_text_embedder", FastembedTextEmbedder(
     model="sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2", prefix="Đại diện cho câu này để tìm kiếm các đoạn văn có liên quan: ")
     )
    querying.add_component("retriever", QdrantHybridRetriever(document_store=document_store))
    querying.add_component("document_joiner", DocumentJoiner())
    querying.add_component("ranker", TransformersSimilarityRanker(model="BAAI/bge-m3"))    
    querying.add_component("prompt_builder", prompt_builder)
    querying.add_component("llm", generator)
    
    querying.connect("sparse_text_embedder.sparse_embedding", "retriever.query_sparse_embedding")
    querying.connect("dense_text_embedder.embedding", "retriever.query_embedding")
    querying.connect("retriever", "document_joiner")
    querying.connect("document_joiner", "ranker")
    querying.connect("ranker.documents", "prompt_builder.documents")
    querying.connect("prompt_builder", "llm")
    querying.debug=True
    results = querying.run(
        {
            "dense_text_embedder": {"text": prompt},
            "sparse_text_embedder": {"text": prompt},
            "ranker": {"query": prompt},
            "prompt_builder": {"question": prompt}
        }
    )


    end_time = time.time()

    elapsed_time = end_time - start_time

    print(f"Execution time: {elapsed_time:.6f} seconds")
    
    return results

@app.get("/download-database/")
async def download_database():
    import time

    start_time = time.time()
    
    # Path to the database directory
    database_dir = join(os.getcwd(), 'database')
    # Path for the zip file
    zip_path = join(os.getcwd(), 'database.zip')
    
    # Create a zip file of the database directory
    shutil.make_archive(zip_path.replace('.zip', ''), 'zip', database_dir)

    end_time = time.time()

    elapsed_time = end_time - start_time

    print(f"Execution time: {elapsed_time:.6f} seconds")
    
    # Return the zip file as a response for download
    return FileResponse(zip_path, media_type='application/zip', filename='database.zip')

def truncate_text(text: str) -> str:
    if len(text) <= 3000:
        return text
    else:
        return text[:3000]
        
@app.post("/query2metadata/")
async def extract_metadata_from_query(query: str):
    from QueryMetadataExtractor import QueryMetadataExtractor
    
    extractor = QueryMetadataExtractor()
    metadata_fields = {"publisher", "publish_date", "document_type"}
    
    return extractor.run(query, metadata_fields)
    
@app.post("/pdf2text/")
async def convert_upload_file(file: UploadFile = File(...)):
    import pytesseract
    from pdf2image import convert_from_path
    from octoai.client import OctoAI
    from octoai.text_gen import ChatCompletionResponseFormat, ChatMessage
    
    file_savePath =  join(temp_path,file.filename)

    with open(file_savePath,'wb') as f:
        shutil.copyfileobj(file.file, f)
    # convert PDF to image
    images = convert_from_path(file_savePath)

    text=""
    first_page = ""

    # Extract text from images
    for image in images:
        ocr_text = pytesseract.image_to_string(image,lang='vie')
        if first_page=="":
            first_page = truncate_text(ocr_text)
        text=text+ocr_text+'\n'
        
    client = OctoAI()

    completion = client.text_gen.create_chat_completion(
        model="meta-llama-3-70b-instruct",
        messages=[
            ChatMessage(role="system", content="You are a helpful assistant."),
            ChatMessage(role="user", content=first_page),
        ],
        presence_penalty=0,
        temperature=0.1,
        top_p=0.9,
        response_format=ChatCompletionResponseFormat(
            type="json_object",
            schema=Invoice.model_json_schema(),
        ),
    )
    
    return {'content':text,'metadate':completion.choices[0].message.content}
    
def image2metadata(image):
    try:
        image = image.convert('RGB')
        answer = model.chat(
            image=image,
            msgs=msgs,
            tokenizer=tokenizer,
            **params
        )
        for char in answer:
            yield char
    except Exception as err:
        print(err)
        traceback.print_exc()
        yield ERROR_MSG
        
@app.post("/pdf2metadata/")
async def convert_upload_file2(file: UploadFile = File(...)):
    import pytesseract
    
    from pdf2image import convert_from_path
    
    file_savePath =  join(temp_path,file.filename)

    with open(file_savePath,'wb') as f:
        shutil.copyfileobj(file.file, f)
    # convert PDF to image
    images = convert_from_path(file_savePath)

    text=""
    first_page = ""

    # Extract text from images
    for image in images:
        question="Sử dụng tiếng Việt để trích thông tin từ hóa đơn sau đó trả ra dưới dạng JSON, Trong bảng chi tiết hóa đơn bỏ qua dòng có các ô [A,B,C,1,2,3=1x2]"
        context = [{"role": "user", "content": _question}]
        params = {
                'sampling': False,
                'stream': False,
                'num_beams': num_beams,
                'repetition_penalty': repetition_penalty,
                "max_new_tokens": 896 
            }
        ocr_text = chat(image, context, None, params=params, vision_hidden_states=None)
        #ocr_text = pytesseract.image_to_string(image,lang='vie')
        if first_page=="":
            first_page = chat(ocr_text)
        text=text+ocr_text+'\n'
        
    client = OctoAI()

    completion = client.text_gen.create_chat_completion(
        model="meta-llama-3-70b-instruct",
        messages=[
            ChatMessage(role="system", content="You are a helpful assistant."),
            ChatMessage(role="user", content=first_page),
        ],
        presence_penalty=0,
        temperature=0.1,
        top_p=0.9,
        response_format=ChatCompletionResponseFormat(
            type="json_object",
            schema=Invoice.model_json_schema(),
        ),
    )
    
    return {'content':text,'metadate':completion.choices[0].message.content}

    
@app.get("/")
def api_home():
    return {'detail': 'Welcome to FastAPI Qdrant importer!'}