Update app.py

app.py

@@ -33,6 +33,7 @@ def predict(text, seed, out_seq_length, min_gen_length, sampling_strategy,
         "num_beams": num_beams,
         "max_tokens": out_seq_length,
         "no_repeat_ngram": no_repeat_ngram_size,
+        "quantization": "int4",
         "seed": seed
     })