Change use_cache to True which significantly speeds up inference

config.json

@@ -18,6 +18,6 @@
   "tie_word_embeddings": false,
   "torch_dtype": "float16",
   "transformers_version": "4.29.0.dev0",
-  "use_cache": false,
+  "use_cache": true,
   "vocab_size": 32001
 }