Update app.py

app.py

@@ -34,7 +34,7 @@ repetition_penalty=1.7
 tokenizer = transformers.AutoTokenizer.from_pretrained(model_name)
 model = transformers.AutoModelForCausalLM.from_pretrained(model_name,
     device_map="auto",
-    torch_dtype=torch.bfloat16,
+    torch_dtype=torch.float16,
 #   load_in_4bit=True
 )
 model.eval()