Update app.py

Add `bfloat16` support for lighter (maybe faster too?) inference. I used to add this argument on `pipeline`, see for example https://gist.github.com/younesbelkada/dba25f75d3749b4e2d2d4821f0d6f385#file-benchmark-py-L42 /

app.py

@@ -1,5 +1,6 @@
 import os
 import gradio as gr
+import torch
 import numpy as np
 from transformers import pipeline
 
@@ -7,8 +8,8 @@ import torch
 print(f"Is CUDA available: {torch.cuda.is_available()}")
 print(f"CUDA device: {torch.cuda.get_device_name(torch.cuda.current_device())}")
 
-pipe_flan = pipeline("text2text-generation", model="google/flan-t5-xl", device="cuda:0")
-pipe_vanilla = pipeline("text2text-generation", model="t5-large", device="cuda:0")
+pipe_flan = pipeline("text2text-generation", model="google/flan-t5-xl", device="cuda:0", model_kwargs={"torch_dtype":torch.bfloat16})
+pipe_vanilla = pipeline("text2text-generation", model="t5-large", device="cuda:0", model_kwargs={"torch_dtype":torch.bfloat16})
 
 examples = [
   ["Please answer to the following question. Who is going to be the next Ballon d'or?"],