default tokenizer_type should be string

For BPE tokenization, `tokenizer_type` argument should be a string instead of the list.

megatron/tokenizer/train_tokenizer.py

@@ -102,7 +102,7 @@ def parse_args():
  type=str,
  help="type of tokenizer to train, currently only BPE is supported",
  choices=["BPE"],
- default=["BPE"],
+ default="BPE",
  )
  parser.add_argument(
  "-v",