Update app.py

app.py

@@ -1,18 +1,45 @@
-import gradio as gr
+
+import huggingface_hub as hf_hub
 import time
+import openvino_genai as ov_genai
+import numpy as np
+import gradio as gr
+import re
+
+# 下載模型
+model_id = "OpenVINO/Qwen3-0.6B-int4-ov"
+model_path = "Qwen3-0.6B-int4-ov"
+
+hf_hub.snapshot_download(model_id, local_dir=model_path, local_dir_use_symlinks=False)
+
+# 建立推理管線
+device = "CPU"
+pipe = ov_genai.LLMPipeline(model_path, device)
+tokenizer = pipe.get_tokenizer()
+tokenizer.set_chat_template(tokenizer.chat_template)
+
 
-def generate_markdown():
-    response = ""
-    for word in ["Hello", "World", "Gradio", "Markdown", "!"]:
-        response += word + " "
-        time.sleep(0.5)
-        yield response
+def generate_response(prompt):
+    try:
+        generated = pipe.generate([prompt], max_length=1024)
+        tokenpersec=f'{generated.perf_metrics.get_throughput().mean:.2f}'
 
-with gr.Blocks() as demo:
-    markdown_output = gr.Markdown(label="回應")
-    button = gr.Button("生成")
+        return tokenpersec, generated
+    except Exception as e:
+        return "發生錯誤", "發生錯誤", f"生成回應時發生錯誤：{e}"
+    
 
-    button.click(generate_markdown, outputs=markdown_output)
+# 建立 Gradio 介面
+demo = gr.Interface(
+    fn=generate_response,
+    inputs=gr.Textbox(lines=5, label="輸入提示 (Prompt)"),
+    outputs=[
+        gr.Textbox(label="tokens/sec"),
+        gr.Textbox(label="回應")
+    ],
+    title="Qwen3-0.6B-int4-ov ",
+    description="基於 Qwen3-0.6B-int4-ov 推理應用，支援思考過程分離與 GUI。"
+)
 
-demo.queue()
-demo.launch()
+if __name__ == "__main__":
+    demo.launch()