Update app.py

app.py

@@ -1,39 +1,29 @@
 import gradio as gr
-from transformers import AutoTokenizer, AutoModelForCausalLM, AutoImageProcessor
-import torch
-from PIL import Image
-import os
+from transformers import AutoTokenizer, AutoModelForCausalLM, AutoProcessor
 from huggingface_hub import login
+import os
+import torch
 
-# ✅ 登入 Token（注意，不要寫死 token，請用 Secrets）
 HF_TOKEN = os.environ.get("HF_TOKEN")
 login(token=HF_TOKEN)
 
-# ✅ 模型與處理器
-MODEL_ID = "Qwen/Qwen-VL-Chat"
+MODEL_ID = "Qwen/Qwen-VL-Chat-Int4"
 
-tokenizer = AutoTokenizer.from_pretrained(MODEL_ID, trust_remote_code=True, token=HF_TOKEN)
-image_processor = AutoImageProcessor.from_pretrained(MODEL_ID, trust_remote_code=True, token=HF_TOKEN)
-model = AutoModelForCausalLM.from_pretrained(MODEL_ID, trust_remote_code=True, token=HF_TOKEN).eval()
+processor = AutoProcessor.from_pretrained(MODEL_ID, trust_remote_code=True, token=HF_TOKEN)
+model = AutoModelForCausalLM.from_pretrained(MODEL_ID, trust_remote_code=True, device_map="auto", token=HF_TOKEN)
+model.eval()
 
-# ✅ 推理函數
 def ask(image, prompt):
-    image_tensor = image_processor(image, return_tensors="pt")["pixel_values"].to(model.device)
-    text_input = tokenizer(prompt, return_tensors="pt").to(model.device)
-    inputs = {
-        "input_ids": text_input["input_ids"],
-        "pixel_values": image_tensor
-    }
-    output = model.generate(**inputs, max_new_tokens=512)
-    response = tokenizer.decode(output[0], skip_special_tokens=True)
+    inputs = processor(images=image, text=prompt, return_tensors="pt").to(model.device)
+    outputs = model.generate(**inputs, max_new_tokens=512)
+    response = processor.batch_decode(outputs, skip_special_tokens=True)[0]
     return response
 
-# ✅ Gradio UI
 demo = gr.Interface(
     fn=ask,
     inputs=[gr.Image(type="pil"), gr.Textbox(label="請輸入問題")],
     outputs="text",
-    title="🧠 Qwen-VL 圖文問答 Demo"
+    title="🧠 Qwen1.5-VL 圖文問答 Demo"
 )
 
 if __name__ == "__main__":