Update app.py

app.py

@@ -17,8 +17,8 @@ def greet(text):
     batch = tokenizer(f"'{text}' ->: ", return_tensors='pt')
 
     # Use torch.no_grad to disable gradient calculation
-    with torch.no_grad():
-        output_tokens = model.generate(**batch, do_sample=True, max_new_tokens=20)
+    with torch.cuda.amp.autocast():
+        output_tokens = model.generate(**batch, max_new_tokens=20)
 
     return tokenizer.decode(output_tokens[0], skip_special_tokens=True)