Fixed tokenizer_config.json

Fixed tokenizer_config.json's inconsistent BOS, UNK token contents. Also added a `padding_side` configuration as same as SFT's tokenizer.

tokenizer_config.json

@@ -3,7 +3,7 @@
   "add_prefix_space": false,
   "bos_token": {
     "__type": "AddedToken",
-    "content": "<|endoftext|>",
+    "content": "<||bos||>",
     "lstrip": false,
     "normalized": true,
     "rstrip": false,
@@ -21,10 +21,11 @@
   "errors": "replace",
   "model_max_length": 8192,
   "pad_token": null,
+  "padding_side": "right",
   "tokenizer_class": "GPT2Tokenizer",
   "unk_token": {
     "__type": "AddedToken",
-    "content": "<|endoftext|>",
+    "content": "<||unk||>",
     "lstrip": false,
     "normalized": true,
     "rstrip": false,