app.py

import os
import numpy as np
import cv2
import pytesseract
from pdf2image import convert_from_path
from fastapi import FastAPI, UploadFile, File
from transformers import AutoTokenizer, AutoModelForCausalLM

app = FastAPI()

# Hugging Face LLMの設定
MODEL_NAME = "deepseek-ai/deepseek-coder-1.3b"
tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
model = AutoModelForCausalLM.from_pretrained(MODEL_NAME)

# OCR関数（決算短信PDF -> テキスト変換）
def extract_text_from_pdf(pdf_path: str) -> str:
    images = convert_from_path(pdf_path)
    extracted_text = ""

    for image in images:
        img_cv = cv2.cvtColor(np.array(image), cv2.COLOR_RGB2BGR)
        gray = cv2.cvtColor(img_cv, cv2.COLOR_BGR2GRAY)
        thresh = cv2.threshold(gray, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)[1]
        text = pytesseract.image_to_string(thresh, lang="jpn")
        extracted_text += text + "\n"

    return extracted_text

# LLM要約関数（決算短信のOCRテキスト -> 投資家向け要約）
def summarize_text(text: str) -> str:
    prompt = f"以下の決算短信を投資家向けに要約してください:\n{text}"
    inputs = tokenizer(prompt, return_tensors="pt", truncation=True, max_length=4096)
    output_ids = model.generate(inputs.input_ids, max_length=512, num_beams=5, early_stopping=True)
    return tokenizer.decode(output_ids[0], skip_special_tokens=True)

# FastAPIエンドポイント（ファイルアップロード & OCR処理 & LLM要約）
@app.post("/upload/")
async def upload_pdf(file: UploadFile = File(...)):
    file_path = f"/tmp/{file.filename}"
    
    # ファイルを保存
    with open(file_path, "wb") as f:
        f.write(await file.read())

    # OCRでテキスト抽出
    extracted_text = extract_text_from_pdf(file_path)

    # LLMで要約
    summary = summarize_text(extracted_text)

    return {"summary": summary}