layoutlmv3-base-chinese tokenizer could not be loaded.

[wlhgtc]

The resources in chinese version layoutlmv3 is sentencepiece.bpe.model . But we seem need vocab.json and merges.txt to load the LayoutLMv3Tokenizer . So @Dod-o could you provide a function to convert them or confirm whether there is a diff between these two tokenizers?

[Sanster]

layoutlmv3-base-chinese use XLMRobertaTokenizer as tokenizer, add "tokenizer_class": "XLMRobertaTokenizer" in https://huggingface.co/microsoft/layoutlmv3-base-chinese/blob/main/config.json works for me.

[pogevip]

Same problem. How to solve it?

[pogevip]

layoutlmv3-base-chinese 使用 XLMRobertaTokenizer 作为标记器，添加https://huggingface.co/microsoft/layoutlmv3-base-chinese/blob/main/config.json"tokenizer_class": "XLMRobertaTokenizer"对我有用。

what is your transformers version? transformers newest not work.

[wlhgtc]

@HYPJUDY Seem the XLMRobertaTokenizer define in config file only accept text as input; but the LayoutLMv3Tokenizer accept both text and bboxes as input. So it could add bboxes info to tokenized res automatic. It would be more convenient.

[Sanster]

layoutlmv3-base-chinese 使用 XLMRobertaTokenizer 作为标记器，添加[https://huggingface.co/microsoft/layoutlmv3-base-chinese/blob/main/config.json"tokenizer_class](https://huggingface.co/microsoft/layoutlmv3-base-chinese/blob/main/config.json%60%22tokenizer_class)": "XLMRobertaTokenizer"对我有用。

what is your transformers version? transformers newest not work.

4.5.1

[Dod-o]

Hi @wlhgtc, as mentioned above, the Chinese version of layoutlmv3 uses XLMRobertaTokenizer, it is the difference between layoutlmv3-zh and layoutlmv3-en. This code provides an example of how to process bbox by ourselves.

[wlhgtc]

Hi @wlhgtc, as mentioned above, the Chinese version of layoutlmv3 uses XLMRobertaTokenizer, it is the difference between layoutlmv3-zh and layoutlmv3-en. This code provides an example of how to process bbox by ourselves.

Actually, I create a new tokenzier from LayoutLMv3Tokenizer, use spm model to tokenize text, rather than the two files.

SPIECE_UNDERLINE = "▁"

class LayoutLMv3ChineseTokenizer(LayoutLMv3Tokenizer):
    def __init__(
            self,
            vocab_file,
            merges_file,
            errors="replace",
            bos_token="<s>",
            eos_token="</s>",
            sep_token="</s>",
            cls_token="<s>",
            unk_token="<unk>",
            pad_token="<pad>",
            mask_token="<mask>",
            add_prefix_space=True,
            cls_token_box=[0, 0, 0, 0],
            sep_token_box=[0, 0, 0, 0],
            pad_token_box=[0, 0, 0, 0],
            pad_token_label=-100,
            only_label_first_subword=True,
            sp_model_kwargs: Optional[Dict[str, Any]] = None,
            **kwargs
    ):
        bos_token = AddedToken(bos_token, lstrip=False, rstrip=False) if isinstance(bos_token, str) else bos_token
        eos_token = AddedToken(eos_token, lstrip=False, rstrip=False) if isinstance(eos_token, str) else eos_token
        sep_token = AddedToken(sep_token, lstrip=False, rstrip=False) if isinstance(sep_token, str) else sep_token
        cls_token = AddedToken(cls_token, lstrip=False, rstrip=False) if isinstance(cls_token, str) else cls_token
        unk_token = AddedToken(unk_token, lstrip=False, rstrip=False) if isinstance(unk_token, str) else unk_token
        pad_token = AddedToken(pad_token, lstrip=False, rstrip=False) if isinstance(pad_token, str) else pad_token

        # Mask token behave like a normal word, i.e. include the space before it
        mask_token = AddedToken(mask_token, lstrip=True, rstrip=False) if isinstance(mask_token, str) else mask_token

        self.sp_model_kwargs = {} if sp_model_kwargs is None else sp_model_kwargs

        super().__init__(
            vocab_file=vocab_file,
            merges_file=merges_file,
            errors=errors,
            bos_token=bos_token,
            eos_token=eos_token,
            unk_token=unk_token,
            sep_token=sep_token,
            cls_token=cls_token,
            pad_token=pad_token,
            mask_token=mask_token,
            add_prefix_space=add_prefix_space,
            cls_token_box=cls_token_box,
            sep_token_box=sep_token_box,
            pad_token_box=pad_token_box,
            pad_token_label=pad_token_label,
            only_label_first_subword=only_label_first_subword,
            sp_model_kwargs=self.sp_model_kwargs,
            **kwargs,
        )

        with open(vocab_file, encoding="utf-8") as vocab_handle:
            self.encoder = json.load(vocab_handle)
        self.decoder = {v: k for k, v in self.encoder.items()}
        self.errors = errors  # how to handle errors in decoding
        self.byte_encoder = bytes_to_unicode()
        self.byte_decoder = {v: k for k, v in self.byte_encoder.items()}

        # with open(merges_file, encoding="utf-8") as merges_handle:
        #     bpe_merges = merges_handle.read().split("\n")[1:-1]
        # bpe_merges = [tuple(merge.split()) for merge in bpe_merges]
        # self.bpe_ranks = dict(zip(bpe_merges, range(len(bpe_merges))))

        sp_model_file = kwargs["name_or_path"] + "/sentencepiece.bpe.model"

        self.sp_model = spm.SentencePieceProcessor(**self.sp_model_kwargs)
        self.sp_model.Load(str(sp_model_file))

        self.cache = {}
        self.add_prefix_space = add_prefix_space

        # Should have added re.IGNORECASE so BPE merges can happen for capitalized versions of contractions
        # self.pat = re.compile(r"""'s|'t|'re|'ve|'m|'ll|'d| ?\p{L}+| ?\p{N}+| ?[^\s\p{L}\p{N}]+|\s+(?!\S)|\s+""")

        # additional properties
        self.cls_token_box = cls_token_box
        self.sep_token_box = sep_token_box
        self.pad_token_box = pad_token_box
        self.pad_token_label = pad_token_label
        self.only_label_first_subword = only_label_first_subword

    def _tokenize(self, text: str) -> List[str]:
        return self.sp_model.encode(text, out_type=str)


    def convert_tokens_to_string(self, tokens):
        """Converts a sequence of tokens (strings for sub-words) in a single string."""
        out_string = "".join(tokens).replace(SPIECE_UNDERLINE, " ").strip()
        return out_string

Maybe you could check whether it's available, or just do some minor change to LayoutLMv3Tokenizer. Then it could fit both languages?
@Dod-o

[YBAgg]

layoutlmv3-base-chinese 使用 XLMRobertaTokenizer 作为标记器，添加[https://huggingface.co/microsoft/layoutlmv3-base-chinese/blob/main/config.json"tokenizer_class](https://huggingface.co/microsoft/layoutlmv3-base-chinese/blob/main/config.json%60%22tokenizer_class)": "XLMRobertaTokenizer"对我有用。

what is your transformers version? transformers newest not work.

have you solve this problem?

[pogevip]

layoutlmv3-base-chinese 使用 XLMRobertaTokenizer 作为标记器，添加[https://huggingface.co/microsoft/layoutlmv3-base-chinese/blob/main/config.json"tokenizer_class](https://huggingface.co/microsoft/layoutlmv3-base-chinese/blob/main/config.json%60%22tokenizer_class)": "XLMRobertaTokenizer"对我有用。

what is your transformers version? transformers newest not work.

have you solve this problem? I tried changing the code, but it didn't work. In the end I use unilm and it's ok. https://github.com/microsoft/unilm/tree/master/layoutlmv3

[aixuedegege]

把这个tokenizer.json放到模型目录下。AutoProcessor load完之后记得用sentencepiece load一下sentencepiece.bpe.model 核对一下processor load出来的对不对

[BarryRun]

Use LayoutLMv3TokenizerFast() instead of LayoutLMv3Tokenizer()