sd

tokenizeConfig.py

@@ -1,222 +1,262 @@
-"""This is an educational implementation of the byte pair encoding algorithm."""
-import collections
-from typing import Optional
+# Copyright (c) 2023, Baichuan Intelligent Technology. All rights reserved.
+
+import os
+from shutil import copyfile
+from typing import Any, Dict, List, Optional, Tuple
+
+import sentencepiece as spm
+from transformers.tokenization_utils import AddedToken, PreTrainedTokenizer
+from transformers.utils import logging
+
+
+logger = logging.get_logger(__name__)
+
+VOCAB_FILES_NAMES = {"vocab_file": "tokenizer.model"}
+
+PRETRAINED_VOCAB_FILES_MAP = {
+    "vocab_file": {},
+    "tokenizer_file": {},
+}
+PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES = {}
+
+
+class BaichuanTokenizer(PreTrainedTokenizer):
+    """
+    Construct a Baichuan tokenizer. Based on byte-level Byte-Pair-Encoding.
+    Args:
+        vocab_file (`str`):
+            Path to the vocabulary file.
+    """
+
+    vocab_files_names = VOCAB_FILES_NAMES
+    pretrained_vocab_files_map = PRETRAINED_VOCAB_FILES_MAP
+    max_model_input_sizes = PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES
+    model_input_names = ["input_ids", "attention_mask"]
+
+    def __init__(
+        self,
+        vocab_file,
+        unk_token="<unk>",
+        bos_token="<s>",
+        eos_token="</s>",
+        pad_token=None,
+        sp_model_kwargs: Optional[Dict[str, Any]] = None,
+        add_bos_token=True,
+        add_eos_token=False,
+        clean_up_tokenization_spaces=False,
+        **kwargs,
+    ):
+        self.sp_model_kwargs = {} if sp_model_kwargs is None else sp_model_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
+        )
+        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
+        )
+        super().__init__(
+            bos_token=bos_token,
+            eos_token=eos_token,
+            unk_token=unk_token,
+            pad_token=pad_token,
+            add_bos_token=add_bos_token,
+            add_eos_token=add_eos_token,
+            sp_model_kwargs=self.sp_model_kwargs,
+            clean_up_tokenization_spaces=clean_up_tokenization_spaces,
+            **kwargs,
+        )
+        self.vocab_file = vocab_file
+        self.add_bos_token = add_bos_token
+        self.add_eos_token = add_eos_token
+        self.sp_model = spm.SentencePieceProcessor(**self.sp_model_kwargs)
+        self.sp_model.Load(vocab_file)
+
+    def __getstate__(self):
+        state = self.__dict__.copy()
+        state["sp_model"] = None
+        return state
+
+    def __setstate__(self, d):
+        self.__dict__ = d
+        self.sp_model = spm.SentencePieceProcessor(**self.sp_model_kwargs)
+        self.sp_model.Load(self.vocab_file)
+
+    @property
+    def vocab_size(self):
+        """Returns vocab size"""
+        return self.sp_model.get_piece_size()
+    
+
+   
+
+    def get_vocab(self):
+        """Returns vocab as a dict"""
+        vocab = {self.convert_ids_to_tokens(i): i for i in range(self.vocab_size)}
+        vocab.update(self.added_tokens_encoder)
+        return vocab
+
+    def _tokenize(self, text):
+        """Returns a tokenized string."""
+        return self.sp_model.encode(text, out_type=str)
+
+    def _convert_token_to_id(self, token):
+        """Converts a token (str) in an id using the vocab."""
+        return self.sp_model.piece_to_id(token)
+
+    def _convert_id_to_token(self, index):
+        """Converts an index (integer) in a token (str) using the vocab."""
+        token = self.sp_model.IdToPiece(index)
+        return token
+
+    def convert_tokens_to_string(self, tokens):
+        """Converts a sequence of tokens (string) in a single string."""
+        current_sub_tokens = []
+        out_string = ""
+        prev_is_special = False
+        for i, token in enumerate(tokens):
+            # make sure that special tokens are not decoded using sentencepiece model
+            if token in self.all_special_tokens:
+                if not prev_is_special and i != 0:
+                    out_string += " "
+                out_string += self.sp_model.decode(current_sub_tokens) + token
+                prev_is_special = True
+                current_sub_tokens = []
+            else:
+                current_sub_tokens.append(token)
+                prev_is_special = False
+        out_string += self.sp_model.decode(current_sub_tokens)
+        return out_string
+    
+    def _encode(self,text):
+        tokens = self._tokenize(text)
+        ids = self._convert_token_to_id(tokens)
+        return ids
+    
+    def _decode(self,ids):
+        tokens = self._convert_id_to_token(ids)
+        text = self.convert_tokens_to_string(tokens)
+        return text
+
+    def save_vocabulary(
+        self, save_directory, filename_prefix: Optional[str] = None
+    ) -> Tuple[str]:
+        """
+        Save the vocabulary and special tokens file to a directory.
+        Args:
+            save_directory (`str`):
+                The directory in which to save the vocabulary.
+        Returns:
+            `Tuple(str)`: Paths to the files saved.
+        """
+        if not os.path.isdir(save_directory):
+            logger.error(f"Vocabulary path ({save_directory}) should be a directory")
+            return
+        out_vocab_file = os.path.join(
+            save_directory,
+            (filename_prefix + "-" if filename_prefix else "")
+            + VOCAB_FILES_NAMES["vocab_file"],
+        )
 
-import regex
+        if os.path.abspath(self.vocab_file) != os.path.abspath(
+            out_vocab_file
+        ) and os.path.isfile(self.vocab_file):
+            copyfile(self.vocab_file, out_vocab_file)
+        elif not os.path.isfile(self.vocab_file):
+            with open(out_vocab_file, "wb") as fi:
+                content_spiece_model = self.sp_model.serialized_model_proto()
+                fi.write(content_spiece_model)
 
-import tiktoken
+        return (out_vocab_file,)
 
+    def build_inputs_with_special_tokens(self, token_ids_0, token_ids_1=None):
+        bos_token_id = [self.bos_token_id] if self.add_bos_token else []
+        eos_token_id = [self.eos_token_id] if self.add_eos_token else []
 
-class OBITokenizer:
-    def __init__(self, *, pat_str: str, mergeable_ranks: dict[bytes, int]) -> None:
-        """Creates an Encoding object."""
-        # A regex pattern string that is used to split the input text
-        self.pat_str = pat_str
-        # A dictionary mapping token bytes to their ranks. The ranks correspond to merge priority
-        self.mergeable_ranks = mergeable_ranks
+        output = bos_token_id + token_ids_0 + eos_token_id
 
-        self._decoder = {token: token_bytes for token_bytes, token in mergeable_ranks.items()}
-        self._pat = regex.compile(pat_str)
+        if token_ids_1 is not None:
+            output = output + bos_token_id + token_ids_1 + eos_token_id
 
-    def encode(self, text: str, visualise: Optional[str] = "colour") -> list[int]:
-        """Encodes a string into tokens.
+        return output
 
-        >>> enc.encode("hello world")
-        [388, 372]
+    def get_special_tokens_mask(
+        self,
+        token_ids_0: List[int],
+        token_ids_1: Optional[List[int]] = None,
+        already_has_special_tokens: bool = False,
+    ) -> List[int]:
         """
-        # Use the regex to split the text into (approximately) words
-        words = self._pat.findall(text)
-        tokens = []
-        for word in words:
-            # Turn each word into tokens, using the byte pair encoding algorithm
-            word_bytes = word.encode("utf-8")
-            word_tokens = bpe_encode(self.mergeable_ranks, word_bytes, visualise=visualise)
-            tokens.extend(word_tokens)
-        return tokens
-
-    def decode_bytes(self, tokens: list[int]) -> bytes:
-        """Decodes a list of tokens into bytes.
-
-        >>> enc.decode_bytes([388, 372])
-        b'hello world'
+        Retrieve sequence ids from a token list that has no special tokens added. This method is called when adding
+        special tokens using the tokenizer `prepare_for_model` method.
+        Args:
+            token_ids_0 (`List[int]`):
+                List of IDs.
+            token_ids_1 (`List[int]`, *optional*):
+                Optional second list of IDs for sequence pairs.
+            already_has_special_tokens (`bool`, *optional*, defaults to `False`):
+                Whether or not the token list is already formatted with special tokens for the model.
+        Returns:
+            `List[int]`: A list of integers in the range [0, 1]: 1 for a special token, 0 for a sequence token.
         """
-        return b"".join(self._decoder[token] for token in tokens)
-
-    def decode(self, tokens: list[int]) -> str:
-        """Decodes a list of tokens into a string.
-
-        Decoded bytes are not guaranteed to be valid UTF-8. In that case, we replace
-        the invalid bytes with the replacement character "�".
+        if already_has_special_tokens:
+            return super().get_special_tokens_mask(
+                token_ids_0=token_ids_0,
+                token_ids_1=token_ids_1,
+                already_has_special_tokens=True,
+            )
+
+        bos_token_id = [1] if self.add_bos_token else []
+        eos_token_id = [1] if self.add_eos_token else []
+
+        if token_ids_1 is None:
+            return bos_token_id + ([0] * len(token_ids_0)) + eos_token_id
+        return (
+            bos_token_id
+            + ([0] * len(token_ids_0))
+            + eos_token_id
+            + bos_token_id
+            + ([0] * len(token_ids_1))
+            + eos_token_id
+        )
 
-        >>> enc.decode([388, 372])
-        'hello world'
+    def create_token_type_ids_from_sequences(
+        self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None
+    ) -> List[int]:
         """
-        return self.decode_bytes(tokens).decode("utf-8", errors="replace")
-
-    def decode_tokens_bytes(self, tokens: list[int]) -> list[bytes]:
-        """Decodes a list of tokens into a list of bytes.
-
-        Useful for visualising how a string is tokenised.
-
-        >>> enc.decode_tokens_bytes([388, 372])
-        [b'hello', b' world']
+        Creates a mask from the two sequences passed to be used in a sequence-pair classification task. An ALBERT
+        sequence pair mask has the following format:
+        ```
+        0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1
+        | first sequence    | second sequence |
+        ```
+        if token_ids_1 is None, only returns the first portion of the mask (0s).
+        Args:
+            token_ids_0 (`List[int]`):
+                List of ids.
+            token_ids_1 (`List[int]`, *optional*):
+                Optional second list of IDs for sequence pairs.
+        Returns:
+            `List[int]`: List of [token type IDs](../glossary#token-type-ids) according to the given sequence(s).
         """
-        return [self._decoder[token] for token in tokens]
-
-    @staticmethod
-    def train(training_data: str, vocab_size: int, pat_str: str):
-        """Train a BPE tokeniser on some data!"""
-        mergeable_ranks = bpe_train(data=training_data, vocab_size=vocab_size, pat_str=pat_str)
-        return OBITokenizer(pat_str=pat_str, mergeable_ranks=mergeable_ranks)
-
-    @staticmethod
-    def from_tiktoken(encoding):
-        if isinstance(encoding, str):
-            encoding = tiktoken.get_encoding(encoding)
-        return OBITokenizer(
-            pat_str=encoding._pat_str, mergeable_ranks=encoding._mergeable_ranks
-        )
+        bos_token_id = [self.bos_token_id] if self.add_bos_token else []
+        eos_token_id = [self.eos_token_id] if self.add_eos_token else []
+
+        output = [0] * len(bos_token_id + token_ids_0 + eos_token_id)
 
+        if token_ids_1 is not None:
+            output += [1] * len(bos_token_id + token_ids_1 + eos_token_id)
 
-def bpe_encode(
-    mergeable_ranks: dict[bytes, int], input: bytes, visualise: Optional[str] = "colour"
-) -> list[int]:
-    parts = [bytes([b]) for b in input]
-    while True:
-        # See the intermediate merges play out!
-        if visualise:
-            if visualise in ["colour", "color"]:
-                visualise_tokens(parts)
-            elif visualise == "simple":
-                print(parts)
-
-        # Iterate over all pairs and find the pair we want to merge the most
-        min_idx = None
-        min_rank = None
-        for i, pair in enumerate(zip(parts[:-1], parts[1:])):
-            rank = mergeable_ranks.get(pair[0] + pair[1])
-
-            if rank is not None and (min_rank is None or rank < min_rank):
-                min_idx = i
-                min_rank = rank
-
-        # If there were no pairs we could merge, we're done!
-        if min_rank is None:
-            break
-        assert min_idx is not None
-
-        # Otherwise, merge that pair and leave the rest unchanged. Then repeat.
-        parts = parts[:min_idx] + [parts[min_idx] + parts[min_idx + 1]] + parts[min_idx + 2 :]
-
-    if visualise:
-        print()
-
-    tokens = [mergeable_ranks[part] for part in parts]
-    return tokens
-
-
-def bpe_train(
-    data: str, vocab_size: int, pat_str: str, visualise: Optional[str] = "colour"
-) -> dict[bytes, int]:
-    # First, add tokens for each individual byte value
-    if vocab_size < 2**8:
-        raise ValueError("vocab_size must be at least 256, so we can encode all bytes")
-    ranks = {}
-    for i in range(2**8):
-        ranks[bytes([i])] = i
-
-    # Splinter up our data into lists of bytes
-    # data = "Hello world"
-    # words = [
-    #     [b'H', b'e', b'l', b'l', b'o'],
-    #     [b' ', b'w', b'o', b'r', b'l', b'd']
-    # ]
-    words: list[list[bytes]] = [
-        [bytes([b]) for b in word.encode("utf-8")] for word in regex.findall(pat_str, data)
-    ]
-
-    # Now, use our data to figure out which merges we should make
-    while len(ranks) < vocab_size:
-        # Find the most common pair. This will become our next token
-        stats = collections.Counter()
-        for piece in words:
-            for pair in zip(piece[:-1], piece[1:]):
-                stats[pair] += 1
-
-        most_common_pair = max(stats, key=lambda x: stats[x])
-        token_bytes = most_common_pair[0] + most_common_pair[1]
-        token = len(ranks)
-        # Add the new token!
-        ranks[token_bytes] = token
-
-        # Now merge that most common pair in all the words. That is, update our training data
-        # to reflect our decision to make that pair into a new token.
-        new_words = []
-        for word in words:
-            new_word = []
-            i = 0
-            while i < len(word) - 1:
-                if (word[i], word[i + 1]) == most_common_pair:
-                    # We found our pair! Merge it
-                    new_word.append(token_bytes)
-                    i += 2
-                else:
-                    new_word.append(word[i])
-                    i += 1
-            if i == len(word) - 1:
-                new_word.append(word[i])
-            new_words.append(new_word)
-        words = new_words
-
-        # See the intermediate merges play out!
-        if visualise:
-            print(f"The current most common pair is {most_common_pair[0]} + {most_common_pair[1]}")
-            print(f"So we made {token_bytes} our {len(ranks)}th token")
-            if visualise in ["colour", "color"]:
-                print("Now the first fifty words in our training data look like:")
-                visualise_tokens([token for word in words[:50] for token in word])
-            elif visualise == "simple":
-                print("Now the first twenty words in our training data look like:")
-                for word in words[:20]:
-                    print(word)
-            print("\n")
-
-    return ranks
-
-
-def visualise_tokens(token_values: list[bytes]) -> None:
-    background = [f"\u001b[48;5;{i}m" for i in [167, 179, 185, 77, 80, 68, 134]]
-    # If token boundaries do not occur at unicode character boundaries, it's unclear how best to
-    # visualise the token. Here, we'll just use the unicode replacement character to represent some
-    # fraction of a character.
-    unicode_token_values = [x.decode("utf-8", errors="replace") for x in token_values]
-
-    running_length = 0
-    last_color = None
-    for token in unicode_token_values:
-        color = background[running_length % len(background)]
-        if color == last_color:
-            color = background[(running_length + 1) % len(background)]
-            assert color != last_color
-        last_color = color
-        running_length += len(token)
-        print(color + token, end="")
-    print("\u001b[0m")
-
-
-def train_simple_encoding():
-    gpt2_pattern = (
-        r"""'s|'t|'re|'ve|'m|'ll|'d| ?[\p{L}]+| ?[\p{N}]+| ?[^\s\p{L}\p{N}]+|\s+(?!\S)|\s+"""
-    )
-    with open(__file__, "r") as f:
-        data = f.read()
-
-    enc = OBITokenizer.train(data, vocab_size=600, pat_str=gpt2_pattern)
-
-    print("This is the sequence of merges performed in order to encode 'hello world':")
-    tokens = enc.encode("hello world")
-    assert enc.decode(tokens) == "hello world"
-    assert enc.decode_bytes(tokens) == b"hello world"
-    assert enc.decode_tokens_bytes(tokens) == [b"hello", b" world"]
-
-    return enc
+        return output