Update app.py

app.py

@@ -8,15 +8,28 @@ from typing import Iterable
 from gradio.themes import Soft
 from gradio.themes.utils import colors, fonts, sizes
 
-# --- Mock Spaces (保持不变) ---
-class MockSpaces:
-    def GPU(self, duration=0):
+try:
+    import spaces
+except ImportError:
+    class MockSpaces:
+        def GPU(self, duration=0):
+            def decorator(func):
+                return func
+            return decorator
+    spaces = MockSpaces()
+
+if torch.cuda.is_available():
+    print("🚀 RunPod/Local GPU detected: Bypassing Hugging Face Spaces queue.")
+    def gpu_bypass_decorator(duration=0):
         def decorator(func):
             return func
         return decorator
-spaces = MockSpaces()
+    spaces.GPU = gpu_bypass_decorator
+else:
+    print("🐢 No GPU detected: Using standard Spaces logic (or Build Mode).")
+
+# ----------------------------------------
 
-# --- Theme Setup (保持不变) ---
 colors.steel_blue = colors.Color(
     name="steel_blue",
     c50="#EBF3F8",
@@ -84,21 +97,17 @@ class SteelBlueTheme(Soft):
         )
 steel_blue_theme = SteelBlueTheme()
 
-# --- 关键修改：按需加载 ---
 from diffusers import FlowMatchEulerDiscreteScheduler
 from qwenimage.pipeline_qwenimage_edit_plus import QwenImageEditPlusPipeline
 from qwenimage.transformer_qwenimage import QwenImageTransformer2DModel
 from qwenimage.qwen_fa3_processor import QwenDoubleStreamAttnProcessorFA3
 
-pipe = None # 全局变量初始化为空
+pipe = None 
 
-# 检测逻辑：只有在真正有 GPU 的时候才加载模型
-# 这样 HF 的 CPU 构建服务器会直接跳过这里，瞬间完成构建
 if torch.cuda.is_available():
-    print("GPU detected! Initializing model for 2x A40 Environment...")
+    print("🚀 GPU detected! Initializing model for RunPod Environment...")
     dtype = torch.bfloat16
     
-    # 1. Load Transformer (device_map="auto" for multi-gpu split)
     print("Loading Transformer...")
     transformer_model = QwenImageTransformer2DModel.from_pretrained(
         "linoyts/Qwen-Image-Edit-Rapid-AIO", 
@@ -107,7 +116,7 @@ if torch.cuda.is_available():
         device_map="auto" 
     )
 
-    # 2. Load Pipeline (device_map="balanced" compatible with diffusers)
+    # 2. Load Pipeline (device_map="balanced")
     print("Loading Pipeline...")
     pipe = QwenImageEditPlusPipeline.from_pretrained(
         "Qwen/Qwen-Image-Edit-2509",
@@ -132,7 +141,7 @@ if torch.cuda.is_available():
     except Exception as e:
         print(f"Warning: FA3 set skipped: {e}")
 else:
-    print("No GPU detected (likely HF Build Environment). SKIPPING MODEL LOAD to save memory.")
+    print("🐢 No GPU detected (HF Build Environment). SKIPPING MODEL LOAD.")
 
 MAX_SEED = np.iinfo(np.int32).max
 
@@ -154,9 +163,8 @@ def update_dimensions_on_upload(image):
 
 @spaces.GPU(duration=30)
 def infer(input_image, prompt, lora_adapter, seed, randomize_seed, guidance_scale, steps, progress=gr.Progress(track_tqdm=True)):
-    # 运行时检查：如果 pipe 没加载（说明没 GPU），直接报错
     if pipe is None:
-        raise gr.Error("Error: Model not loaded. Is a GPU available?")
+        raise gr.Error("Model not loaded. Are you running on GPU?")
 
     if input_image is None:
         raise gr.Error("Please upload an image to edit.")
@@ -210,7 +218,7 @@ css="""
 
 with gr.Blocks(css=css, theme=steel_blue_theme) as demo:
     with gr.Column(elem_id="col-container"):
-        gr.Markdown("# **Qwen-Image-Edit-2509-LoRAs-Fast (RunPod Optimized)**", elem_id="main-title")
+        gr.Markdown("# **Qwen-Image-Edit-2509 (2x A40 Ready)**", elem_id="main-title")
         
         with gr.Row(equal_height=True):
             with gr.Column():