Update app.py

app.py

@@ -1,70 +1,61 @@
 import gradio as gr
 from transformers import pipeline
 from PIL import Image
+import torch
 
-print("--- 程式開始執行 ---")
+# --- 延後載入：先設為 None，第一次呼叫功能時才載入 ---
+_captioner = None
+_vqa = None
 
-# 1. 載入模型
-# ---------------------------------------------------
-print("開始載入 Image Captioning 模型...")
-captioner = pipeline(
-    "image-to-text", 
-    model="nlpconnect/vit-gpt2-image-captioning"
-)
-print("Image Captioning 模型載入完成。")
+def get_captioner():
+    global _captioner
+    if _captioner is None:
+        _captioner = pipeline(
+            "image-to-text",
+            model="nlpconnect/vit-gpt2-image-captioning",
+            device=0 if torch.cuda.is_available() else -1,
+        )
+    return _captioner
 
-print("開始載入 VQA 模型...")
-vqa = pipeline(
-    "visual-question-answering", 
-    model="dandelin/vilt-b32-finetuned-vqa"
-)
-print("VQA 模型載入完成。")
+def get_vqa():
+    global _vqa
+    if _vqa is None:
+        _vqa = pipeline(
+            "visual-question-answering",
+            model="dandelin/vilt-b32-finetuned-vqa",
+            device=0 if torch.cuda.is_available() else -1,
+        )
+    return _vqa
 
-
-# 2. 定義 AI 功能的函式
-# ---------------------------------------------------
-def image_captioning(image):
-    """
-    看圖說故事
-    """
+# --- 功能 ---
+def image_captioning(image: Image.Image):
     if image is None:
         return "請先上傳圖片"
-    print("正在執行 image_captioning...")
-    result = captioner(image)
-    print("Captioning 結果:", result[0]['generated_text'])
-    return result[0]['generated_text']
+    pipe = get_captioner()
+    out = pipe(image)
+    return out[0]["generated_text"]
 
-def visual_question_answer(image, question):
-    """
-    視覺問答
-    """
+def visual_question_answer(image: Image.Image, question: str):
     if image is None:
         return "請先上傳圖片"
     if not question:
         return "請輸入問題"
-        
-    print(f"正在執行 VQA... 問題: {question}")
-    result = vqa(image, question)
-    print("VQA 結果:", result[0]['answer'])
-    return result[0]['answer']
+    pipe = get_vqa()
+    out = pipe(image, question)
+    return out[0]["answer"]
 
-# 3. 建立 Gradio 介面
-# ---------------------------------------------------
-print("開始建立 Gradio 介面...")
+# --- UI ---
 with gr.Blocks(title="Vision Language AI 應用") as demo:
     gr.Markdown("# 🎨 Vision Language AI 應用展示")
     gr.Markdown("這是一個展示 Vision-Language (VL) 模型的 Web UI，由 Gradio 和 Hugging Face Transformers 驅動。")
 
     with gr.Tabs():
-        
-        # --- 分頁 1: Image Captioning ---
         with gr.TabItem("🖼️ 看圖說故事 (Image Captioning)"):
             with gr.Row():
                 image_input_caption = gr.Image(type="pil", label="上傳圖片")
                 text_output_caption = gr.Textbox(label="AI 產生的描述")
             caption_button = gr.Button("開始產生描述")
 
-        # --- 分頁 2: Visual Question Answering ---
         with gr.TabItem("❓ 視覺問答 (VQA)"):
             with gr.Row():
                 image_input_vqa = gr.Image(type="pil", label="上傳圖片")
@@ -73,59 +64,12 @@ with gr.Blocks(title="Vision Language AI 應用") as demo:
                     text_output_vqa = gr.Textbox(label="AI 的回答")
             vqa_button = gr.Button("開始問答")
 
-    # 4. 綁定按鈕和函式 (***解決 No API found 的關鍵***)
-    # ---------------------------------------------------
-    
-    print("開始綁定 '看圖說故事' 按鈕...")
-    caption_button.click(
-        fn=image_captioning,
-        inputs=[image_input_caption],
-        outputs=[text_output_caption]
-    )
-    
-    print("開始綁定 '視覺問答' 按鈕...")
-    vqa_button.click(
-        fn=visual_question_answer,
-        inputs=[image_input_vqa, question_input_vqa],
-        outputs=[text_output_vqa]
-    )
-    
-    print("按鈕綁定完成。")
-
-    # ---------------------------------------------------
-    # 5. 範例區 (***暫時停用來避開 TypeError Bug***)
-    # ---------------------------------------------------
-    """
-    print("正在處理範例區...") # 這行不會被印出
-    gr.Markdown("--- \n ### 範例圖片")
-    gr.Examples(
-        examples=[
-            ["sample_images/cat.jpeg"],
-            ["sample_images/human.jpg"],
-            ["sample_images/clown.png"],
-        ],
-        inputs=[image_input_caption],
-        outputs=[text_output_caption],
-        fn=image_captioning,
-        cache_examples=True
-    )
-    gr.Examples(
-        examples=[
-            ["sample_images/cat.jpeg", "What color is the cat?"],
-            ["sample_images/human.jpg", "What is the gender of the human?"]
-        ],
-        inputs=[image_input_vqa, question_input_vqa],
-        outputs=[text_output_vqa],
-        fn=visual_question_answer,
-        cache_examples=True
-    )
-    """
+    caption_button.click(image_captioning, inputs=[image_input_caption], outputs=[text_output_caption])
+    vqa_button.click(visual_question_answer, inputs=[image_input_vqa, question_input_vqa], outputs=[text_output_vqa])
 
-print("Gradio 介面建立完成。")
+# ★ 在 Spaces 讓後端自動啟動；本地端才需要 launch()
+# demo.queue(concurrency_count=1, max_size=10) 放在全域即可
+demo.queue(concurrency_count=1, max_size=10)
 
-# 6. 啟動 Web App
-# ---------------------------------------------------
-if __name__ == "__main__":
-    print("準備啟動 App...")
-    demo.launch()
-    print("App 應該已經啟動。")
+if __name__ == "__main__":  # 只在本地 run 時才啟動
+    demo.launch()