Upload app.py

app.py

@@ -1,39 +1,32 @@
 # app.py
 import os, sys
-os.environ.setdefault("HF_HUB_ENABLE_HF_TRANSFER", "1")  # 更快的下载
 import gradio as gr
 import torch
 
+os.environ.setdefault("HF_HUB_ENABLE_HF_TRANSFER", "1")
 MODEL_REPO = "vibevoice/VibeVoice-1.5B"
 
-demo = None
+demo_app = None
 model_ready = False
 
-def ensure_loaded():
-    global demo, model_ready
-    if model_ready:
-        return
-    # 只有第一次真正用到时才加载（避免冷启动卡住健康检查）
-    from demo.gradio_demo import VibeVoiceDemo  # 延迟导入
-    device = "cuda" if torch.cuda.is_available() else "cpu"
-    # 你的 gradio_demo.py 里如何传参就怎么来，这里举例：
-    demo = VibeVoiceDemo(model_path=MODEL_REPO, device=device)  # 内部会 .load_model()
-    model_ready = True
+def load_real_demo():
+    global demo_app, model_ready
+    if not model_ready:
+        from demo import gradio_demo
+        # 假设 gradio_demo.py 里有 main(return_app=True) 可以返回一个 Blocks
+        demo_app = gradio_demo.main(model_path=MODEL_REPO, return_app=True)
+        model_ready = True
+    return demo_app
 
-def warmup():
-    ensure_loaded()
-    return "✅ Model is loaded and ready."
+def launch_wrapper():
+    load_real_demo()
+    return demo_app
 
-def ui():
-    # 你原本的 UI 如果在 gradio_demo.py 里已有 main()，也可以直接 import 后调用。
-    # 这里放一个简单门面，点击后再加载再转到真正的 demo。
-    with gr.Blocks() as app:
-        gr.Markdown("## VibeVoice Demo — click below to start (lazy load)")
-        btn = gr.Button("Start / Load model")
-        out = gr.Textbox(label="Status")
-        btn.click(fn=warmup, outputs=out)
-    return app
+with gr.Blocks() as app:
+    gr.Markdown("## 🎙️ VibeVoice Demo (lazy-load)")
+    btn = gr.Button("🚀 Load full demo")
+    out = gr.HTML()
+    btn.click(fn=lambda: "<b>Reload page to see full demo</b>", outputs=out)
 
 if __name__ == "__main__":
-    app = ui()
     app.queue().launch(server_name="0.0.0.0", server_port=7860)