set use_cache to true

as discussed in this [thread](https://huggingface.co/HuggingFaceH4/starchat-alpha/discussions/3#645fde7625a4075bcf9d2e15) this should make inference much faster

config.json

@@ -33,7 +33,7 @@
   "summary_use_proj": true,
   "torch_dtype": "float16",
   "transformers_version": "4.28.1",
-  "use_cache": false,
+  "use_cache": true,
   "validate_runner_input": true,
   "vocab_size": 49156
 }