Update app.py

app.py

@@ -347,6 +347,8 @@ def generate(text_list, image_list, max_new_tokens, force_boi, force_bbox, force
     generated_text = output['text']
     generated_text = generated_text.replace(EOI_TOKEN, IMG_FLAG).replace(service.tokenizer.eos_token, '')
 
+    torch.cuda.empty_cache() 
+      
     if output['has_img_output']:
         # print('loading visual encoder and llm to CPU, and sd to GPU')
         # a = time.time()
@@ -375,11 +377,10 @@ def generate(text_list, image_list, max_new_tokens, force_boi, force_bbox, force
                 image_base64 = encode_image(generated_image)
                 gen_imgs_base64_list.append(image_base64)
 
-                
                 # service.vae_pipe = service.vae_pipe.to("cpu")
                 # service.sd_adapter = service.sd_adapter.to(service.vit_sd_device, dtype=service.dtype)
                 
-            torch.cuda.empty_cache() 
+                torch.cuda.empty_cache() 
 
         # print('loading visual encoder and llm to GPU, and sd to CPU')
         # a = time.time()