Update app.py

app.py

@@ -101,7 +101,7 @@ def load_image(image_file, input_size=448, max_num=12):
 
 model = AutoModel.from_pretrained(
     "5CD-AI/Vintern-1B-v3_5",
-    torch_dtype=torch.float32,
+    torch_dtype=torch.bfloat16,
     low_cpu_mem_usage=True,
     trust_remote_code=True,
 ).eval().cuda()
@@ -122,12 +122,12 @@ We currently only support one image at the start of the context! Please start a
             test_image = message["files"][0]["path"]
         else:
             test_image = message["files"][0]
-        pixel_values = load_image(test_image, max_num=6).cuda()
+        pixel_values = load_image(test_image, max_num=6).to(torch.bfloat16).cuda()
     elif len(history) == 0 and len(message["files"]) == 0:
         pixel_values = None     
     elif history[0][0][0] is not None and os.path.isfile(history[0][0][0]):
         test_image = history[0][0][0]
-        pixel_values = load_image(test_image, max_num=6).cuda()
+        pixel_values = load_image(test_image, max_num=6).to(torch.bfloat16).cuda()
     else:
         pixel_values = None