app.py

import os
import sys
import tempfile
import gradio as gr
import requests
import uvicorn
import torch
import base64

from fastapi import FastAPI, File, UploadFile
from fastapi.responses import RedirectResponse, StreamingResponse
from typing import List
from pdf2image import convert_from_bytes
from PIL import Image
from torch.utils.data import DataLoader
from transformers import AutoProcessor
from reportlab.pdfgen import canvas
from reportlab.lib.pagesizes import letter
from io import BytesIO

sys.path.append(os.path.abspath(os.path.join(os.path.dirname(__file__), './colpali-main')))

from colpali_engine.models.paligemma_colbert_architecture import ColPali
from colpali_engine.trainer.retrieval_evaluator import CustomEvaluator
from colpali_engine.utils.colpali_processing_utils import (
    process_images,
    process_queries,
)

app = FastAPI()

# Load model
model_name = "vidore/colpali"
token = os.environ.get("HF_TOKEN")
model = ColPali.from_pretrained(
    "google/paligemma-3b-mix-448", torch_dtype=torch.bfloat16, device_map="cpu", token = token).eval()

model.load_adapter(model_name)
processor = AutoProcessor.from_pretrained(model_name, token = token)
device = "cuda:0" if torch.cuda.is_available() else "cpu"
if device != model.device:
    model.to(device)
mock_image = Image.new("RGB", (448, 448), (255, 255, 255))

# In-memory storage
ds = []
images = []

@app.post("/index")
async def index(files: List[UploadFile] = File(...)):
    global ds, images
    images = []
    ds = []
    for file in files:
        content = await file.read()
        pdf_image_list = convert_from_bytes(content)
        images.extend(pdf_image_list)
    
    dataloader = DataLoader(
        images,
        batch_size=4,
        shuffle=False,
        collate_fn=lambda x: process_images(processor, x),
    )
    for batch_doc in dataloader:
        with torch.no_grad():
            batch_doc = {k: v.to(device) for k, v in batch_doc.items()}
            embeddings_doc = model(**batch_doc)
        ds.extend(list(torch.unbind(embeddings_doc.to("cpu"))))
    
    return {"message": f"Uploaded and converted {len(images)} pages"}

@app.post("/search")
async def search(query: str, k: int):
    qs = []
    with torch.no_grad():
        batch_query = process_queries(processor, [query], mock_image)
        batch_query = {k: v.to(device) for k, v in batch_query.items()}
        embeddings_query = model(**batch_query)
        qs.extend(list(torch.unbind(embeddings_query.to("cpu"))))

    retriever_evaluator = CustomEvaluator(is_multi_vector=True)
    scores = retriever_evaluator.evaluate(qs, ds)

    top_k_indices = scores.argsort(axis=1)[0][-k:][::-1]

    results = []
    for idx in top_k_indices:
        img_byte_arr = BytesIO()
        images[idx].save(img_byte_arr, format='PNG')
        img_base64 = base64.b64encode(img_byte_arr.getvalue()).decode('utf-8')
        results.append({"image": img_base64, "page": f"Page {idx}"})

    # Generate PDF
    pdf_buffer = BytesIO()
    c = canvas.Canvas(pdf_buffer, pagesize=letter)
    width, height = letter

    for result in results:
        img_base64 = result["image"]
        img_data = base64.b64decode(img_base64)
        
        # Create a temporary file to hold the image
        with tempfile.NamedTemporaryFile(suffix=".png", delete=False) as temp_file:
            temp_file.write(img_data)
            temp_file.flush()
            
            # Draw the image from the temporary file
            c.drawImage(temp_file.name, 0, 0, width, height)
            c.showPage()
        
        # Clean up the temporary file
        os.remove(temp_file.name)

    c.save()
    pdf_buffer.seek(0)

    # Use StreamingResponse to handle in-memory file
    response = StreamingResponse(pdf_buffer, media_type='application/pdf')
    response.headers['Content-Disposition'] = 'attachment; filename="search_results.pdf"'

    return response
    
if __name__ == "__main__":
    uvicorn.run(app, host="0.0.0.0", port=7860)