Salesforce

/

xgen-7b-8k-base

like 319

Follow

Salesforce 724

Hi team ,

There is an major issue with the tokenizer.get_vocab() method, as get_vocab() is a very important method that is used by a lot of codebases and libraries..for example ctranslate2

Having an issue over here renders the model useless for many usecases

Below are my logs and a possible fix

Thanks,

How to reproduce?
tokenizer = AutoTokenizer.from_pretrained("Salesforce/xgen-7b-8k-base",trust_remote_code=True)
tokenizer.get_vocab()

Possible Fix?
def _convert_id_to_token(self, index):
"""Converts an index (integer) in a token (str) using the vocab."""

•    return self.encoder.decode_single_token_bytes(index).decode("utf-8")
  

•   return self.encoder.decode_single_token_bytes(index).decode("latin1")
  

Use latin1 encoding over here instead of "utf-8"

Error thrown?

UnicodeDecodeError Traceback (most recent call last)
Cell In[39], line 1
----> 1 tokenizer.get_vocab()["<|tiktoken|>"]

File ~/.cache/huggingface/modules/transformers_modules/Salesforce/xgen-7b-8k-base/3987e094377fae577bba039af1b300ee8086f9e1/tokenization_xgen.py:142, in XgenTokenizer.get_vocab(self)
140 def get_vocab(self):
141 """Returns vocab as a dict"""
--> 142 vocab = {self._convert_id_to_token(i): i for i in range(self.vocab_size)}
143 return vocab

File ~/.cache/huggingface/modules/transformers_modules/Salesforce/xgen-7b-8k-base/3987e094377fae577bba039af1b300ee8086f9e1/tokenization_xgen.py:142, in (.0)
140 def get_vocab(self):
141 """Returns vocab as a dict"""
--> 142 vocab = {self._convert_id_to_token(i): i for i in range(self.vocab_size)}
143 return vocab

File ~/.cache/huggingface/modules/transformers_modules/Salesforce/xgen-7b-8k-base/3987e094377fae577bba039af1b300ee8086f9e1/tokenization_xgen.py:158, in XgenTokenizer._convert_id_to_token(self, index)
156 def _convert_id_to_token(self, index):
157 """Converts an index (integer) in a token (str) using the vocab."""
--> 158 return self.encoder.decode_single_token_bytes(index).decode("utf-8")

UnicodeDecodeError: 'utf-8' codec can't decode byte 0xa1 in position 0: invalid start byte