Update app.py

app.py

@@ -61,7 +61,7 @@ model, image_processor, tokenizer = create_model_and_transforms(
 
 checkpoint_path = hf_hub_download("openflamingo/OpenFlamingo-9B-vitl-mpt7b", "checkpoint.pt")
 model.load_state_dict(torch.load(checkpoint_path), strict=False)
-model.eval().to("cuda")
+model.eval().to("cuda", dtype=torch.bfloat16)
 
 def generate(
     idx,
@@ -152,7 +152,7 @@ def generate(
 
     # with torch.cuda.amp.autocast(dtype=torch.bfloat16):
     output = model.generate(
-        vision_x=vision_x.to("cuda"),
+        vision_x=vision_x.to("cuda", dtype=torch.bfloat16)
         lang_x=input_ids.to("cuda"),
         attention_mask=attention_mask.to("cuda"),
         max_new_tokens=30,