tokenization_chatglm.py

import json
import os
import warnings
from typing import Dict, List, Optional, Tuple, Union

from sentencepiece import SentencePieceProcessor
from tokenizers import AddedToken, decoders, normalizers, processors
from transformers import PreTrainedTokenizer, PreTrainedTokenizerFast
from transformers.convert_slow_tokenizer import SLOW_TO_FAST_CONVERTERS, SpmConverter
from transformers.tokenization_utils_base import (
    BatchEncoding,
    EncodedInput,
    PreTokenizedInput,
    PreTokenizedInputPair,
    TextInput,
    TextInputPair,
    TruncationStrategy,
)
from transformers.utils import PaddingStrategy

ADDITIONAL_SPECIAL_TOKENS = [
    "[MASK]",
    "[gMASK]",
    "[sMASK]",
    "<!sop!>",
    "<!eop!>",
    "<|system|>",
    "<|user|>",
    "<|assistant|>",
    "<|observation|>",
]
PREFIX_TOKENS = ["[gMASK]", "<!sop!>"]

ENCODE_SEP_TOKEN_FOR_FAST = "<!encode-sep!>"


class SPTokenizer:
    def __init__(self, model_path: str):
        # reload tokenizer
        assert os.path.isfile(model_path), model_path
        self.sp_model = SentencePieceProcessor(model_file=model_path)

        # BOS / EOS token IDs
        self.n_words: int = self.sp_model.vocab_size()
        self.bos_id: int = self.sp_model.bos_id()
        self.eos_id: int = self.sp_model.eos_id()
        self.pad_id: int = self.sp_model.unk_id()
        assert self.sp_model.vocab_size() == self.sp_model.get_piece_size()

        self.special_tokens = {}
        self.index_special_tokens = {}
        for token in ADDITIONAL_SPECIAL_TOKENS:
            self.special_tokens[token] = self.n_words
            self.index_special_tokens[self.n_words] = token
            self.n_words += 1

    def tokenize(self, s: str):
        return self.sp_model.EncodeAsPieces(s)

    def encode(self, s: str, bos: bool = False, eos: bool = False) -> List[int]:
        assert type(s) is str
        t = self.sp_model.encode(s)
        if bos:
            t = [self.bos_id] + t
        if eos:
            t = t + [self.eos_id]
        return t

    def decode(self, t: List[int]) -> str:
        text, buffer = "", []
        for token in t:
            if token in self.index_special_tokens:
                if buffer:
                    text += self.sp_model.decode(buffer)
                    buffer = []
                text += self.index_special_tokens[token]
            else:
                buffer.append(token)
        if buffer:
            text += self.sp_model.decode(buffer)
        return text

    def decode_tokens(self, tokens: List[str]) -> str:
        text = self.sp_model.DecodePieces(tokens)
        return text

    def convert_token_to_id(self, token):
        """ Converts a token (str) in an id using the vocab. """
        if token in self.special_tokens:
            return self.special_tokens[token]
        return self.sp_model.PieceToId(token)

    def convert_id_to_token(self, index):
        """Converts an index (integer) in a token (str) using the vocab."""
        if index in self.index_special_tokens:
            return self.index_special_tokens[index]
        if index in [self.eos_id, self.bos_id, self.pad_id] or index < 0:
            return ""
        return self.sp_model.IdToPiece(index)


class ChatGLMTokenizer(PreTrainedTokenizer):
    vocab_files_names = {"vocab_file": "tokenizer.model"}

    model_input_names = ["input_ids", "attention_mask", "position_ids"]

    def __init__(self, vocab_file, padding_side="left", clean_up_tokenization_spaces=False, **kwargs):
        self.name = "GLMTokenizer"

        self.vocab_file = vocab_file
        self.tokenizer = SPTokenizer(vocab_file)
        self.special_tokens = {
            "<bos>": self.tokenizer.bos_id,
            "<eos>": self.tokenizer.eos_id,
            "<pad>": self.tokenizer.pad_id
        }
        super().__init__(padding_side=padding_side, clean_up_tokenization_spaces=clean_up_tokenization_spaces, **kwargs)

    def get_command(self, token):
        if token in self.special_tokens:
            return self.special_tokens[token]
        assert token in self.tokenizer.special_tokens, f"{token} is not a special token for {self.name}"
        return self.tokenizer.special_tokens[token]

    @property
    def unk_token(self) -> str:
        return "<unk>"

    @property
    def pad_token(self) -> str:
        return "<unk>"

    @property
    def pad_token_id(self):
        return self.get_command("<pad>")

    @property
    def eos_token(self) -> str:
        return "</s>"

    @property
    def eos_token_id(self):
        return self.get_command("<eos>")

    @property
    def vocab_size(self):
        return self.tokenizer.n_words

    def get_vocab(self):
        """ Returns vocab as a dict """
        vocab = {self._convert_id_to_token(i): i for i in range(self.vocab_size)}
        vocab.update(self.added_tokens_encoder)
        return vocab

    def _tokenize(self, text, **kwargs):
        return self.tokenizer.tokenize(text)

    def _convert_token_to_id(self, token):
        """ Converts a token (str) in an id using the vocab. """
        return self.tokenizer.convert_token_to_id(token)

    def _convert_id_to_token(self, index):
        """Converts an index (integer) in a token (str) using the vocab."""
        return self.tokenizer.convert_id_to_token(index)

    def convert_tokens_to_string(self, tokens: List[str]) -> str:
        return self.tokenizer.decode_tokens(tokens)

    def save_vocabulary(self, save_directory, filename_prefix=None):
        """
        Save the vocabulary and special tokens file to a directory.

        Args:
            save_directory (`str`):
                The directory in which to save the vocabulary.
            filename_prefix (`str`, *optional*):
                An optional prefix to add to the named of the saved files.

        Returns:
            `Tuple(str)`: Paths to the files saved.
        """
        if os.path.isdir(save_directory):
            vocab_file = os.path.join(
                save_directory, self.vocab_files_names["vocab_file"]
            )
        else:
            vocab_file = save_directory

        with open(self.vocab_file, 'rb') as fin:
            proto_str = fin.read()

        with open(vocab_file, "wb") as writer:
            writer.write(proto_str)

        return (vocab_file,)

    def get_prefix_tokens(self):
        return list(map(self.get_command, PREFIX_TOKENS))

    def build_single_message(self, role, metadata, message):
        assert role in ["system", "user", "assistant", "observation"], role
        role_tokens = [self.get_command(f"<|{role}|>")] + self.tokenizer.encode(f"{metadata}\n")
        message_tokens = self.tokenizer.encode(message)
        tokens = role_tokens + message_tokens
        return tokens

    def build_chat_input(self, query, history=None, role="user"):
        if history is None:
            history = []
        input_ids = []
        for item in history:
            content = item["content"]
            if item["role"] == "system" and "tools" in item:
                content = content + "\n" + json.dumps(item["tools"], indent=4, ensure_ascii=False)
            input_ids.extend(self.build_single_message(item["role"], item.get("metadata", ""), content))
        input_ids.extend(self.build_single_message(role, "", query))
        input_ids.extend([self.get_command("<|assistant|>")])
        return self.batch_encode_plus([input_ids], return_tensors="pt", is_split_into_words=True)

    def build_inputs_with_special_tokens(
            self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None
    ) -> List[int]:
        """
        Build model inputs from a sequence or a pair of sequence for sequence classification tasks by concatenating and
        adding special tokens. A BERT sequence has the following format:

        - single sequence: `[CLS] X [SEP]`
        - pair of sequences: `[CLS] A [SEP] B [SEP]`

        Args:
            token_ids_0 (`List[int]`):
                List of IDs to which the special tokens will be added.
            token_ids_1 (`List[int]`, *optional*):
                Optional second list of IDs for sequence pairs.

        Returns:
            `List[int]`: List of [input IDs](../glossary#input-ids) with the appropriate special tokens.
        """
        prefix_tokens = self.get_prefix_tokens()
        token_ids_0 = prefix_tokens + token_ids_0
        if token_ids_1 is not None:
            token_ids_0 = token_ids_0 + token_ids_1 + [self.get_command("<eos>")]
        return token_ids_0

    def _pad(
            self,
            encoded_inputs: Union[Dict[str, EncodedInput], BatchEncoding],
            max_length: Optional[int] = None,
            padding_strategy: PaddingStrategy = PaddingStrategy.DO_NOT_PAD,
            pad_to_multiple_of: Optional[int] = None,
            return_attention_mask: Optional[bool] = None,
    ) -> dict:
        """
        Pad encoded inputs (on left/right and up to predefined length or max length in the batch)

        Args:
            encoded_inputs:
                Dictionary of tokenized inputs (`List[int]`) or batch of tokenized inputs (`List[List[int]]`).
            max_length: maximum length of the returned list and optionally padding length (see below).
                Will truncate by taking into account the special tokens.
            padding_strategy: PaddingStrategy to use for padding.

                - PaddingStrategy.LONGEST Pad to the longest sequence in the batch
                - PaddingStrategy.MAX_LENGTH: Pad to the max length (default)
                - PaddingStrategy.DO_NOT_PAD: Do not pad
                The tokenizer padding sides are defined in self.padding_side:

                    - 'left': pads on the left of the sequences
                    - 'right': pads on the right of the sequences
            pad_to_multiple_of: (optional) Integer if set will pad the sequence to a multiple of the provided value.
                This is especially useful to enable the use of Tensor Core on NVIDIA hardware with compute capability
                `>= 7.5` (Volta).
            return_attention_mask:
                (optional) Set to False to avoid returning attention mask (default: set to model specifics)
        """
        # Load from model defaults
        assert self.padding_side == "left"

        required_input = encoded_inputs[self.model_input_names[0]]
        seq_length = len(required_input)

        if padding_strategy == PaddingStrategy.LONGEST:
            max_length = len(required_input)

        if max_length is not None and pad_to_multiple_of is not None and (max_length % pad_to_multiple_of != 0):
            max_length = ((max_length // pad_to_multiple_of) + 1) * pad_to_multiple_of

        needs_to_be_padded = padding_strategy != PaddingStrategy.DO_NOT_PAD and len(required_input) != max_length

        # Initialize attention mask if not present.
        if "attention_mask" not in encoded_inputs:
            encoded_inputs["attention_mask"] = [1] * seq_length

        if "position_ids" not in encoded_inputs:
            encoded_inputs["position_ids"] = list(range(seq_length))

        if needs_to_be_padded:
            difference = max_length - len(required_input)

            if "attention_mask" in encoded_inputs:
                encoded_inputs["attention_mask"] = [0] * difference + encoded_inputs["attention_mask"]
            if "position_ids" in encoded_inputs:
                encoded_inputs["position_ids"] = [0] * difference + encoded_inputs["position_ids"]
            encoded_inputs[self.model_input_names[0]] = [self.pad_token_id] * difference + required_input

        return encoded_inputs


class ChatGLMTokenizerFast(PreTrainedTokenizerFast):
    # multiple breaking changes, no more backward-compatibility
    slow_tokenizer_class = ChatGLMTokenizer
    vocab_files_names = {
        **ChatGLMTokenizer.vocab_files_names,
        **PreTrainedTokenizerFast.vocab_files_names,
    }

    def __init__(self, **kwargs):
        kwargs.setdefault("clean_up_tokenization_spaces", False)
        kwargs.setdefault("bos_token", "<s>")
        kwargs.setdefault("eos_token", "</s>")
        kwargs.setdefault("unk_token", "<unk>")
        kwargs.setdefault("pad_token", "<unk>")
        super().__init__(**kwargs)

    @property
    def encode_sep_token(self):
        return ENCODE_SEP_TOKEN_FOR_FAST

    def _batch_encode_plus(
        self,
        batch_text_or_text_pairs: Union[
            List[TextInput],
            List[TextInputPair],
            List[PreTokenizedInput],
            List[PreTokenizedInputPair],
        ],
        add_special_tokens: bool = True,
        padding_strategy: PaddingStrategy = PaddingStrategy.DO_NOT_PAD,
        truncation_strategy: TruncationStrategy = TruncationStrategy.DO_NOT_TRUNCATE,
        max_length: Optional[int] = None,
        stride: int = 0,
        is_split_into_words: bool = False,
        pad_to_multiple_of: Optional[int] = None,
        return_tensors: Optional[str] = None,
        return_token_type_ids: Optional[bool] = None,
        return_attention_mask: Optional[bool] = None,
        return_overflowing_tokens: bool = False,
        return_special_tokens_mask: bool = False,
        return_offsets_mapping: bool = False,
        return_length: bool = False,
        verbose: bool = True,
    ) -> BatchEncoding:
        def split_sep(t: Union[TextInput, PreTokenizedInput]) -> PreTokenizedInput:
            if isinstance(t, str):
                return t.split(self.encode_sep_token)

            return [w for word in t for w in split_sep(word)]

        def split_maybe_tupled(
            t: Union[TextInput, TextInputPair, PreTokenizedInput, PreTokenizedInputPair]
        ) -> Union[PreTokenizedInputPair, PreTokenizedInput]:
            if isinstance(t, tuple):
                return split_sep(t[0]), split_sep(t[1])

            return split_sep(t)

        return super()._batch_encode_plus(
            list(map(split_maybe_tupled, batch_text_or_text_pairs)),  # pyright: ignore
            add_special_tokens,
            padding_strategy,
            truncation_strategy,
            max_length,
            stride,
            True,
            pad_to_multiple_of,
            return_tensors,
            return_token_type_ids,
            return_attention_mask,
            return_overflowing_tokens,
            return_special_tokens_mask,
            return_offsets_mapping,
            return_length,
            verbose,
        )

    @property
    def can_save_slow_tokenizer(self) -> bool:
        # multiple breaking changes
        return False

    def save_pretrained(
        self,
        save_directory: Union[str, os.PathLike],
        legacy_format: Optional[bool] = None,
        filename_prefix: Optional[str] = None,
        push_to_hub: bool = False,
        **kwargs,
    ) -> Tuple[str]:
        warnings.warn(
            f"{type(self)} does not support saving slow tokenizer. "
            "Saving it at the same directory may break the slow tokenizer. "
            "Please keep a backup of the original tokenizer beforehand."
        )
        return super().save_pretrained(
            save_directory, legacy_format, filename_prefix, push_to_hub, **kwargs
        )

    def build_single_message(self, role, metadata, message):
        assert role in ["system", "user", "assistant", "observation"], role
        return f"<|{role}|>{self.encode_sep_token}{metadata}\n{self.encode_sep_token}{message}"

    def build_chat_text(self, query, history=None, role="user", metadata=""):
        inputs = []

        for item in history or []:
            content = item["content"]

            if item["role"] == "system" and "tools" in item:
                content += "\n" + json.dumps(
                    item["tools"], indent=4, ensure_ascii=False
                )

            inputs.append(
                self.build_single_message(
                    item["role"], item.get("metadata", ""), content
                )
            )

        inputs.append(self.build_single_message(role, metadata, query))
        inputs.append("<|assistant|>")

        return "".join(inputs)

    def build_chat_input(self, *args, **kwargs):
        return self.batch_encode_plus(
            [self.build_chat_text(*args, **kwargs)],
            return_tensors="pt",
        )


ChatGLMTokenizer.register_for_auto_class()
ChatGLMTokenizerFast.register_for_auto_class()


class ChatGLMTokenizerConverter(SpmConverter):
    handle_byte_fallback = True

    def normalizer(self, proto):
        return normalizers.Sequence(
            [
                normalizers.Prepend(prepend="▁"),
                normalizers.Replace(pattern=" ", content="▁"),
            ]
        )

    def pre_tokenizer(self, replacement, add_prefix_space):
        # don't use Metaspace, it will skip merging spaces into one token

        # give up to split `encode_sep_token` here, buggy
        # return pre_tokenizers.Split(ENCODE_SEP_TOKEN_FOR_FAST, "removed")

        return None

    def decoder(self, replacement, add_prefix_space):
        return decoders.Sequence(
            [
                decoders.ByteFallback(),
                super().decoder(replacement, add_prefix_space),
            ]
        )

    def tokenizer(self, proto):
        tokenizer = super().tokenizer(proto)

        tokenizer.model.byte_fallback = True

        special_tokens = [
            "<unk>",
            "<s>",
            "</s>",
            *ADDITIONAL_SPECIAL_TOKENS,
        ]

        tokenizer.add_special_tokens(
            [
                AddedToken(token, special=True, normalized=False)
                for token in special_tokens
            ]
        )

        return tokenizer

    def converted(self):
        tokenizer = super().converted()

        # Post processors
        prefix_token_ids = list(map(tokenizer.token_to_id, PREFIX_TOKENS))
        assert all(i is not None for i in prefix_token_ids)
        prefix_template = " ".join(PREFIX_TOKENS)

        template_special_tokens = list(frozenset(zip(PREFIX_TOKENS, prefix_token_ids)))

        if "</s>" not in PREFIX_TOKENS:
            eos_token_id = tokenizer.token_to_id("</s>")
            assert eos_token_id is not None
            template_special_tokens.append(("</s>", eos_token_id))

        post = processors.TemplateProcessing(
            single=f"{prefix_template} $A",
            pair=f"{prefix_template} $A $B:1 </s>:1",
            special_tokens=template_special_tokens,
        )
        if tokenizer.post_processor is None:
            tokenizer.post_processor = post
        else:
            tokenizer.post_processor = processors.Sequence(
                [tokenizer.post_processor, post]
            )

        return tokenizer


SLOW_TO_FAST_CONVERTERS[ChatGLMTokenizer.__name__] = ChatGLMTokenizerConverter