Update app.py

app.py

@@ -17,8 +17,7 @@ tokenizer = AutoTokenizer.from_pretrained(
 
 model = LlavaQwen2ForCausalLM.from_pretrained(
     'qnguyen3/nanoLLaVA-1.5',
-    torch_dtype=torch.float16,
-    attn_implementation="flash_attention_2",
+    torch_dtype=torch.bfloat16,
     trust_remote_code=True,
     device_map='auto')
 
@@ -84,7 +83,6 @@ def bot_streaming(message, history):
         messages.append({"role": "user", "content": f"<image>\n{message['text']}"})
     elif len(history) == 0 and image is None:
         messages.append({"role": "user", "content": message['text'] })
-    model = model.to('cuda')
 
     # if image is None:
     #     gr.Error("You need to upload an image for LLaVA to work.")