Prevent tokenizer calls from returning `token_type_ids`

Hi there 👋 `transformers` maintainer here.

Adding this extra tokenization config class will prevent the tokenizer from returning `token_type_ids`, which are unused by `SeedOss` 🤗 The default value for this flag is `["input_ids", "token_type_ids", "attention_mask"]`

tokenizer_config.json

@@ -1029,6 +1029,10 @@
   "clean_up_tokenization_spaces": false,
   "eos_token": "<seed:eos>",
   "extra_special_tokens": {},
+  "model_input_names": [
+    "input_ids",
+    "attention_mask"
+  ],
   "model_max_length": 1000000000000000019884624838656,
   "pad_token": "<seed:pad>",
   "tokenizer_class": "PreTrainedTokenizerFast"