turn on use_cache for fast inference

config.json

@@ -33,7 +33,7 @@
   "summary_use_proj": true,
   "torch_dtype": "float16",
   "transformers_version": "4.30.0.dev0",
-  "use_cache": false,
+  "use_cache": true,
   "validate_runner_input": true,
   "vocab_size": 49153
 }