fix app torch grad

app.py

@@ -83,12 +83,11 @@ def text_to_image_generation(input_text, guidance_scale=1.75, generation_timeste
             config=config,
         )
 
-    gen_token_ids = torch.clamp(gen_token_ids, max=config.model.showo.codebook_size - 1, min=0)
-    images = vq_model.decode_code(gen_token_ids)
-
-    images = torch.clamp((images + 1.0) / 2.0, min=0.0, max=1.0)
-    images *= 255.0
-    images = images.permute(0, 2, 3, 1).cpu().numpy().astype(np.uint8)
+        gen_token_ids = torch.clamp(gen_token_ids, max=config.model.showo.codebook_size - 1, min=0)
+        images = vq_model.decode_code(gen_token_ids)
+        images = torch.clamp((images + 1.0) / 2.0, min=0.0, max=1.0)
+        images *= 255.0
+        images = images.permute(0, 2, 3, 1).cpu().detach().numpy().astype(np.uint8)
 
     return images[0]
 
@@ -158,12 +157,12 @@ def text_guided_inpainting(input_text, inpainting_image, inpainting_mask, guidan
             config=config,
         )
 
-    gen_token_ids = torch.clamp(gen_token_ids, max=config.model.showo.codebook_size - 1, min=0)
-    images = vq_model.decode_code(gen_token_ids)
+        gen_token_ids = torch.clamp(gen_token_ids, max=config.model.showo.codebook_size - 1, min=0)
+        images = vq_model.decode_code(gen_token_ids)
 
-    images = torch.clamp((images + 1.0) / 2.0, min=0.0, max=1.0)
-    images *= 255.0
-    images = images.permute(0, 2, 3, 1).cpu().numpy().astype(np.uint8)
+        images = torch.clamp((images + 1.0) / 2.0, min=0.0, max=1.0)
+        images *= 255.0
+        images = images.permute(0, 2, 3, 1).cpu().detach().numpy().astype(np.uint8)
 
     return images[0]
 
@@ -283,11 +282,12 @@ def text_guided_extrapolation(input_img, input_text, left_ext, right_ext, guidan
 
     _, h, w = gen_token_ids.shape
     gen_token_ids = gen_token_ids.reshape(config.training.batch_size, -1)
-    images = vq_model.decode_code(gen_token_ids, shape=(h, w))
 
-    images = torch.clamp((images + 1.0) / 2.0, min=0.0, max=1.0)
-    images *= 255.0
-    images = images.permute(0, 2, 3, 1).cpu().numpy().astype(np.uint8)
+    with torch.no_grad():
+        images = vq_model.decode_code(gen_token_ids, shape=(h, w))
+        images = torch.clamp((images + 1.0) / 2.0, min=0.0, max=1.0)
+        images *= 255.0
+        images = images.permute(0, 2, 3, 1).cpu().detach().numpy().astype(np.uint8)
 
     return images[0]