test

app.py

@@ -3,6 +3,7 @@ import spaces
 from visualcloze import VisualClozeModel
 import gradio as gr
 import examples
+import torch
 from functools import partial
 from data.prefix_instruction import get_layout_instruction
 from huggingface_hub import snapshot_download
@@ -27,7 +28,7 @@ GUIDANCE = """
 
 ## 🔥 Task Examples:
 Click the task button in the right bottom to acquire **examples** of various tasks. 
-Make sure all images and prompts are loaded before clicking the generate button.
+**Make sure all images and prompts are loaded before clicking the generate button.**
 
 
 ## 💻 Runtime on the Zero GPU: 
@@ -338,7 +339,7 @@ def create_demo(model):
                     upsampling_steps=upsampling_steps, upsampling_noise=upsampling_noise
                 )
             except Exception as e:
-                raise gr.Error('Process error. Possible that the task examples have not finished loading yet. Error: ' + e)
+                raise gr.Error('Process error. Possible that the task examples have not finished loading yet. Error: ' + str(e))
 
             output = gr.update(
                 elem_id='output_gallery', 
@@ -502,14 +503,15 @@ def generate(
     prompts, 
     seed, cfg, steps, 
     upsampling_steps, upsampling_noise):
-    return model.process_images(
-        images=images, 
-        prompts=prompts, 
-        seed=seed, 
-        cfg=cfg, 
-        steps=steps, 
-        upsampling_steps=upsampling_steps, 
-        upsampling_noise=upsampling_noise)
+    with torch.no_grad():
+        return model.process_images(
+            images=images, 
+            prompts=prompts, 
+            seed=seed, 
+            cfg=cfg, 
+            steps=steps, 
+            upsampling_steps=upsampling_steps, 
+            upsampling_noise=upsampling_noise)
 
 
 def parse_args():

visualcloze.py

@@ -450,6 +450,7 @@ class VisualClozeModel:
             row_end = grid_h * grid_w
             for i in range(row_start, row_end):
                 # when the image is masked, then output it
+                print(mask_position, i, row_start, is_upsampling, grid_h, grid_w)
                 if mask_position[i - row_start] and is_upsampling:
                     cropped = output_images[-1].crop(((i - row_start) * ret_w // self.grid_w, 0, ((i - row_start) + 1) * ret_w // self.grid_w, ret_h))
                     upsampled = self.upsampling(