Added tokenizer config so the tokenizer correctly loads as cased.

tokenizer_config.json

@@ -0,0 +1,4 @@
+{
+  "do_lower_case": false,
+  "max_len": 512
+}