Update app.py

app.py

@@ -34,14 +34,14 @@ from torchvision import transforms
 from models.controlnet import ControlNetModel
 from models.unet_2d_condition import UNet2DConditionModel
 
-VLM_NAME  = "Qwen/Qwen2.5-VL-3B-Instruct"
+# VLM_NAME  = "Qwen/Qwen2.5-VL-3B-Instruct"
 
-vlm_model = Qwen2_5_VLForConditionalGeneration.from_pretrained(
-    VLM_NAME,
-    torch_dtype="auto",
-    device_map="auto"   # immediately dispatches layers onto available GPUs
-)
-vlm_processor = AutoProcessor.from_pretrained(VLM_NAME)
+# vlm_model = Qwen2_5_VLForConditionalGeneration.from_pretrained(
+#     VLM_NAME,
+#     torch_dtype="auto",
+#     device_map="auto"   # immediately dispatches layers onto available GPUs
+# )
+# vlm_processor = AutoProcessor.from_pretrained(VLM_NAME)
 
 def _generate_vlm_prompt(
     vlm_model: Qwen2_5_VLForConditionalGeneration,
@@ -186,18 +186,20 @@ def process(
     latent_tiled_overlap = 4,
     sample_times = 1,
     ) -> List[np.ndarray]:
+
+    input_image = input_image.resize(256, 256)
     
     process_size = 512
     resize_preproc = transforms.Compose([
         transforms.Resize(process_size, interpolation=transforms.InterpolationMode.BILINEAR),
     ])
-    user_prompt = _generate_vlm_prompt(
-                vlm_model=vlm_model,
-                vlm_processor=vlm_processor,
-                process_vision_info=process_vision_info,
-                pil_image=input_image,
-                device=device,
-            )
+    # user_prompt = _generate_vlm_prompt(
+    #             vlm_model=vlm_model,
+    #             vlm_processor=vlm_processor,
+    #             process_vision_info=process_vision_info,
+    #             pil_image=input_image,
+    #             device=device,
+    #         )
     
     # with torch.no_grad():
     seed_everything(seed)