app.py

import os
import gc
import gradio as gr
from PIL import Image
from transformers import pipeline
import gradio, gradio_client
print("Gradio versions =>", gradio.__version__, "/", gradio_client.__version__)

# 環境變數：減少警告與執行緒開銷
os.environ["TRANSFORMERS_NO_ADVISORY_WARNINGS"] = "true"
os.environ["TOKENIZERS_PARALLELISM"] = "false"
os.environ["PIP_DISABLE_PIP_VERSION_CHECK"] = "1"

# 嘗試偵測 torch（若沒裝也能跑，僅用 CPU）
try:
    import torch
    _HAS_TORCH = True
except Exception:
    torch = None
    _HAS_TORCH = False

# 懶載入
_captioner = None

def get_device_id():
    if _HAS_TORCH:
        try:
            return 0 if torch.cuda.is_available() else -1
        except Exception:
            return -1
    return -1

def get_captioner():
    global _captioner
    if _captioner is None:
        print("📦 載入模型中（BLIP base）...")
        _captioner = pipeline(
            task="image-to-text",
            model="Salesforce/blip-image-captioning-base",
            device=get_device_id(),
        )
        print("✅ 模型載入完成。")
    return _captioner

def do_caption(image: Image.Image):
    if image is None:
        return "請先上傳圖片。"
    pipe = get_captioner()
    out = pipe(image)
    gc.collect()
    return out[0]["generated_text"]

def clear_all():
    return None, ""

# ========= UI =========
with gr.Blocks(title="Vision Language AI 應用（Caption 版）") as demo:
    gr.Markdown("# 🎨 Vision Language AI 應用展示 · Caption")
    gr.Markdown("上傳圖片後按下按鈕，產生文字描述。此版本僅啟用 **Image Captioning**，以確保在免費 CPU 環境穩定執行。")

    with gr.Row():
        img = gr.Image(type="pil", label="上傳圖片", height=320)
        out = gr.Textbox(label="AI 產生的描述", lines=6, show_copy_button=True)

    with gr.Row():
        run_btn = gr.Button("🚀 產生描述", variant="primary")
        clear_btn = gr.Button("🧹 清除", variant="secondary")

    # ⚙️ 事件綁定：放在 Blocks 區塊內
    run_btn.click(fn=do_caption, inputs=img, outputs=out)
    clear_btn.click(fn=clear_all, inputs=None, outputs=[img, out])

demo.queue(max_size=10)

if __name__ == "__main__":
    demo.queue(max_size=10).launch()