tf_text.FastSentencepieceTokenizer causes ValueError: cannot create std::vector larger than max_size()

I tried to switch from the tf_text.SentencepieceTokenizer to tf_text.FastSentencepieceTokenizer but encountered the issue below.

ValueError                                Traceback (most recent call last)
<ipython-input-150-df5232e30d34> in <module>
      7 # sp_model = open("new.model", "r").read()
      8 
----> 9 tokenizer = tf_text.FastSentencepieceTokenizer(sp_model)

[/usr/local/lib/python3.8/dist-packages/tensorflow_text/python/ops/fast_sentencepiece_tokenizer.py](https://localhost:8080/#) in __init__(self, model, reverse, add_bos, add_eos)
     48 
     49   def __init__(self, model, reverse=False, add_bos=False, add_eos=False):
---> 50     converted_model = pywrap_model_converter.convert_sentencepiece_model(model)
     51     converted_model_detokenizer = pywrap_model_converter.convert_sentencepiece_model_for_decoder(
     52         model)

ValueError: cannot create std::vector larger than max_size()

I can load the sentencepiece model proto with tf_text.SentencepieceTokenizer, but switching to tf_text.FastSentencepieceTokenizer is causing the issue.
Is there a limitation around the tf_text.FastSentencepieceTokenizer?

I have tried to trace the C++ code, but didn't find any model limitation.

Start in fast_sentencepiece_tokenizer.py at the constructor and trace the pywrap_model_converter conversion calls shown in the traceback. Reproduce the failure with the SentencePiece model and compare it with SentencepieceTokenizer's loading path. Done means identifying the model or conversion condition that triggers the exception and documenting or correcting the FastSentencepieceTokenizer behavior.