add model_max_length

model_max_length is present in the BERT base tokenizer config, but not here, and it should be

without this, the model_max_length for the tokenizer defaults to a very large int, and inference crashes if not handled properly

tokenizer_config.json

@@ -1,3 +1,4 @@
 {
-  "do_lower_case": true
+  "do_lower_case": true,
+  "model_max_length": 512
 }