Upload tokenizer

added_tokens.json

@@ -0,0 +1,3 @@
+{
+  "<|im_start|>": 250680
+}

gptx_tokenizer.py

@@ -0,0 +1,463 @@
+from __future__ import annotations
+
+import json
+import os
+import warnings
+from pathlib import Path
+from typing import Any, Dict, List, Mapping, Optional, Tuple, Union
+
+import sentencepiece as spm
+import numpy as np
+import torch
+from huggingface_hub import hf_hub_download, list_repo_files, try_to_load_from_cache
+from transformers.tokenization_utils import PreTrainedTokenizer
+from transformers.tokenization_utils_base import TOKENIZER_CONFIG_FILE
+
+
+REPO_ID = "openGPT-X/Teuken-7B-instruct-commercial-v0.4"
+
+class HFGPTXTokenizer(PreTrainedTokenizer):
+    """
+    A custom tokenizer class that extends Hugging Face's PreTrainedTokenizer.
+    It is specifically designed to work with SentencePiece models and integrates
+    with Hugging Face's tokenizer utilities.
+    """
+    
+    model_file_glob = "*tokenizer.json"
+    vocab_files_names = {"tokenizer_file": "tokenizer.json"}
+    decode_kwargs: List[str] = []
+
+    def _encode(self, text: str, return_tokens: bool = False, is_continuation: bool = False):
+        """
+        Encode a given text using the tokenizer.
+        
+        Args:
+            text (str): The text to encode.
+            return_tokens (bool): If True, returns token strings instead of token IDs.
+            is_continuation (bool): If True, uses a continuation tokenizer (if available).
+        Returns:
+            List[int] or List[str]: Encoded text as a list of token IDs or token strings.
+        """
+        assert self.tok is not None, "No tokenizer is currently loaded"
+
+        # Variant with additional sp processor:
+        tokenizer = self.continuation_tokenizer if is_continuation else self.tok
+
+        if return_tokens:
+            return tokenizer.encode_as_pieces(text)
+        else:
+            return tokenizer.encode(text)
+    
+    def create_list_of_special_tokens(self) -> List[str]:
+        """
+        Create a list of special tokens, including the BOS, EOS, PAD, EOD tokens,
+        and 256 additional placeholder tokens.
+        Returns:
+            List[str]: List of special tokens.
+        """
+        return [self.bos_token, self.eos_token, self.pad_token, self.eod_token] + [
+            f"<placeholder_tok_{i}>" for i in range(256)
+        ]
+    
+    def find_tokenizer_config(self, config_path: Path, repo_id: str = None) -> Optional[Path]:
+        if not os.path.isfile(config_path):
+            config_path = try_to_load_from_cache(repo_id=repo_id, filename=Path(config_path).name)
+            if not config_path:
+                config_path = self._download_config_from_hub(repo_id=repo_id)
+
+        return config_path
+
+    
+    def instantiate_from_file_or_name(self, model_file_or_name: str, repo_id: str = None):
+        """
+        Load the tokenizer model from a file or download it from a repository.
+
+        Args:
+            model_file_or_name (str): Path to the model file or the model name.
+            repo_id (str, optional): Repository ID from which to download the model file.
+
+        Returns:
+            spm.SentencePieceProcessor: Loaded SentencePieceProcessor instance.
+
+        Raises:
+            ValueError: If repo_id is not provided when model_file_or_name is not a file.
+            OSError: If the model file cannot be loaded or downloaded.
+        """
+        if not os.path.isfile(model_file_or_name):
+            model_file_or_name = try_to_load_from_cache(repo_id=repo_id, filename=Path(model_file_or_name).name)
+            if not model_file_or_name:
+                model_file_or_name = self._download_model_from_hub(repo_id=repo_id)
+
+        try:
+            return spm.SentencePieceProcessor(model_file=model_file_or_name)
+        except Exception as e:
+            raise OSError(f"Failed to load tokenizer model: {str(e)}")
+
+    def _download_model_from_hub(self, repo_id: str) -> Optional[str]:
+        try:
+            # List all files in the repo
+            repo_files = list_repo_files(repo_id)
+
+            # Find the tokenizer model file
+            tokenizer_files = [f for f in repo_files if f.endswith('.model')]
+            if not tokenizer_files:
+                raise FileNotFoundError(f"No .model file found in repository {repo_id}")
+
+            # Use the first .model file found
+            model_file = tokenizer_files[0]
+            print(f"Found tokenizer model file: {model_file}")
+
+            # Download the file
+            model_file_or_name = hf_hub_download(repo_id=repo_id, filename=model_file)
+            print(f"Downloaded tokenizer model to: {model_file_or_name}")
+        except Exception as e:
+            raise OSError(f"Failed to download tokenizer model: {str(e)}")
+
+        return model_file_or_name
+
+    def _download_config_from_hub(self, repo_id: str):
+        if repo_id is None:
+            raise ValueError("repo_id must be provided if config_path is not a local file")
+
+        try:
+            # List all files in the repo
+            repo_files = list_repo_files(repo_id)
+
+            # Find the tokenizer config file
+            tokenizer_files = [f for f in repo_files if f.endswith('tokenizer_config.json')]
+            if not tokenizer_files:
+                raise FileNotFoundError(f"No tokenizer_config.json file found in repository {repo_id}")
+
+            # Use the first tokenizer_config.json file found
+            tokenizer_config_file = tokenizer_files[0]
+            print(f"Found tokenizer config file: {tokenizer_config_file}")
+
+            # Download the file
+            tokenizer_config_file_or_name = hf_hub_download(repo_id=repo_id, filename=tokenizer_config_file)
+            print(f"Downloaded tokenizer config file to: {tokenizer_config_file_or_name}")
+            return tokenizer_config_file_or_name
+        except Exception as e:
+            raise OSError(f"Failed to download tokenizer model: {str(e)}")    
+    def __init__(
+        self,
+        model_path: Optional[str] = None,
+        config_path: Optional[str] = None,
+        **kwargs: Any,
+    ) -> None:
+        """
+        Initialize the tokenizer.
+        Args:
+            model_path (Optional[str]): Path to the tokenizer model file.
+            config_path (Optional[str]): Path to the tokenizer configuration file.
+            **kwargs: Additional keyword arguments passed to the superclass.
+        This method also ensures backward compatibility by setting
+        `clean_up_tokenization_spaces` to False by default.
+        """
+        # Prevent cleanup of tokenization spaces to maintain backward compatibility
+        self.clean_up_tokenization_spaces = kwargs.setdefault("clean_up_tokenization_spaces", False)
+        self.vocab = None
+        cp_path = kwargs.get("name_or_path", ".")
+        if model_path is None:
+            model_path = str(Path(cp_path) / self.vocab_files_names["tokenizer_file"])
+        self.tok = self.instantiate_from_file_or_name(model_path, repo_id=REPO_ID)
+
+        super().__init__(**kwargs)
+
+        # Specify special tokens which we know the value of.
+        # EOD from `tok` is used as what is called EOS in HuggingFace.
+        # Since there is no corresponding mapping for EOS from `tok` in
+        # HuggingFace, it is treated as an additional special token.
+        # Same for all other special tokens.
+        
+        
+        self.unk_token = "<unk>"
+        self.eos_token = "</s>"
+        self.bos_token = "<s>"
+        self.pad_token = "<pad>"
+        self.eod_token = "<eod>"
+        
+        self.additional_special_tokens = self.create_list_of_special_tokens()
+    
+        if config_path is None:
+            config_path = str(Path(cp_path) / TOKENIZER_CONFIG_FILE)
+
+        if os.path.isfile(config_path):
+            self.tokenizer_config = self.load_json(Path(config_path))
+        else: # Load from repo
+            self.tokenizer_config = self.load_json(Path(self.find_tokenizer_config(Path(config_path), repo_id=REPO_ID)))
+
+    @property
+    def vocab_size(self) -> int:
+        """
+        Get the size of the tokenizer vocabulary.
+        Returns:
+            int: The size of the vocabulary.
+        """
+        return self.tok.GetPieceSize()
+
+    def get_vocab(self) -> Dict[str, int]:
+        """
+        Get the vocabulary as a dictionary mapping token strings to their IDs.
+        Returns:
+            Dict[str, int]: Vocabulary mapping.
+        """
+        if self.vocab is None:
+            self.vocab = {self.tok.IdToPiece(i): i for i in range(self.vocab_size)}
+        return self.vocab
+
+    def _tokenize(self, text: str, **kwargs) -> List[int]:
+        """
+        Tokenize the input text.
+        Args:
+            text (str): Text to tokenize.
+            **kwargs: Additional keyword arguments.
+        Returns:
+            List[int]: List of token IDs.
+        """
+        return_tokens = kwargs.pop("return_tokens", True)
+        return self._encode(text, return_tokens=return_tokens, **kwargs)
+
+    def _convert_token_to_id(self, token: str) -> int:
+        """
+        Convert a token string to its corresponding ID.
+        Args:
+            token (str): The token to convert.
+        Returns:
+            int: The token's ID.
+        Raises:
+            ValueError: If the token is unknown and cannot be encoded to a single ID.
+        """
+        return self.tok.PieceToId(token)
+
+
+    def decode(
+        self,
+        token_ids: Union[List[int], List[List[int]]],
+        num_threads: Optional[int] = None,
+        skip_special_tokens: bool = False,
+        clean_up_tokenization_spaces: bool = False,
+    ) -> str:
+        """
+        Decode a list of token IDs into a string.
+        Args:
+            token_ids (Union[List[int], List[List[int]]]): List of token IDs or lists of token IDs.
+            num_threads (Optional[int]): Number of threads to use for decoding.
+        Returns:
+            str: Decoded string.
+        """
+        if isinstance(token_ids, torch.Tensor):  # For PyTorch tensors
+            token_ids = token_ids.tolist()
+        elif isinstance(token_ids, np.ndarray):  # For NumPy arrays
+            token_ids = token_ids.tolist()
+        
+        output = self.tok.decode(input=token_ids, num_threads=num_threads)
+        if skip_special_tokens:
+            for substring in self.additional_special_tokens:
+                output = output.replace(substring, "")
+        
+        if clean_up_tokenization_spaces:
+            warnings.warn(
+                "when cleaning up tokenization spaces, this will not behave "
+                "like the original `GPTXTokenizer`., Please supply "
+                "`clean_up_tokenization_spaces=False` for decoding."
+            )
+            output = self.clean_up_tokenization(output)
+        
+        return output
+
+    
+    def _convert_id_to_token(self, index: int) -> str:
+        """
+        Convert a token ID to its corresponding token string.
+        Args:
+            index (int): Token ID.
+        Returns:
+            str: Corresponding token string.
+        """
+        return self.tok.IdToPiece(index)
+
+    def convert_tokens_to_string(self, tokens: List[str]) -> str:
+        """
+        Convert a list of tokens into a single string.
+        Args:
+            tokens (List[str]): List of token strings.
+        Returns:
+            str: Concatenated string of tokens.
+        """
+        return self.tok.DecodePieces(tokens)
+
+    def _tok_decode(self, token_ids: List[int], **kwargs: Any) -> str:
+        """
+        Internal method to decode token IDs with additional arguments.
+        Args:
+            token_ids (List[int]): List of token IDs.
+            **kwargs: Additional arguments to pass to the decode method.
+        Returns:
+            str: Decoded string.
+        This method also issues a warning if unsupported arguments are provided.
+        """
+        passed_kwargs = {key: value for (key, value) in kwargs.items() if key in self.decode_kwargs}
+        if len(passed_kwargs) != len(kwargs):
+            warnings.warn("silently ignoring some arguments to `decode` due to missing " "support from the tokenizer.")
+        text = self.decode(token_ids, **passed_kwargs)
+        return text
+    
+    def save_tokenizer(self, save_dir: str) -> None:
+        if not os.path.isdir(save_dir):
+            print(f"Vocabulary path ({save_dir}) should be a directory")
+            return
+        out_vocab_file = os.path.join(save_dir, "tokenizer.model")
+
+        # 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 f:
+            content_spiece_model = self.tok.serialized_model_proto()
+            f.write(content_spiece_model)
+
+        return (out_vocab_file,)
+        
+    def _decode(
+        self,
+        token_ids: List[int],
+        skip_special_tokens: bool = False,
+        clean_up_tokenization_spaces: bool = None,
+        spaces_between_special_tokens: bool = True,
+        **kwargs: Any,
+    ) -> str:
+        text = self._tok_decode(
+            token_ids,
+            skip_special_tokens=skip_special_tokens,
+            spaces_between_special_tokens=spaces_between_special_tokens,
+            **kwargs,
+        )
+
+        clean_up_tokenization_spaces = (
+            clean_up_tokenization_spaces
+            if clean_up_tokenization_spaces is not None
+            else self.clean_up_tokenization_spaces
+        )
+        if clean_up_tokenization_spaces:
+            warnings.warn(
+                "when cleaning up tokenization spaces, this will not behave "
+                "like the original `GPTXTokenizer`., Please supply "
+                "`clean_up_tokenization_spaces=False` for decoding."
+            )
+            clean_text = self.clean_up_tokenization(text)
+            return clean_text
+        else:
+            return text
+        
+    def save_vocabulary(
+        self,
+        save_directory: str,
+        filename_prefix: Optional[str] = None,
+    ) -> Tuple[str]:
+        filename_prefix = filename_prefix + "-" if filename_prefix else ""
+        save_directory = Path(save_directory)
+
+        self._save_tokenizer_config(save_directory, filename_prefix)
+        tokenizer_file_path = self._save_tokenizer(save_directory, filename_prefix)
+
+        return (tokenizer_file_path,)
+
+    def _save_tokenizer_config(
+        self,
+        save_directory: Path,
+        filename_prefix: str,
+    ) -> str:
+        self.save_tokenizer_config(save_directory)
+        old_tokenizer_config_path = save_directory / TOKENIZER_CONFIG_FILE
+        assert old_tokenizer_config_path.is_file(), "tokenizer config path changed"
+        new_tokenizer_config_path = save_directory / (filename_prefix + old_tokenizer_config_path.name)
+        old_tokenizer_config_path.replace(new_tokenizer_config_path)
+        return str(new_tokenizer_config_path)
+
+    def _find_tokenizer_files(self, save_directory: Path) -> List[Path]:
+        files = list(Path(save_directory).glob(self.model_file_glob))
+        return files
+
+    def _get_tokenizer_file(self, files: List[Path]):
+        assert files, "no saved tokenizer file found"
+        assert len(files) <= 1, "cannot handle multiple saved tokenizer files"
+        return files[0]
+    
+    def _save_tokenizer(
+        self,
+        save_directory: Path,
+        filename_prefix: str,
+    ) -> str:
+        self.save_tokenizer(str(save_directory))
+        tokenizer_files = self._find_tokenizer_files(save_directory)
+        old_tokenizer_file_path = self._get_tokenizer_file(tokenizer_files)
+        assert old_tokenizer_file_path.is_file(), "could not access saved tokenizer file"
+        new_tokenizer_file_path = save_directory / (filename_prefix + self.vocab_files_names["tokenizer_file"])
+        old_tokenizer_file_path.replace(new_tokenizer_file_path)
+        return str(new_tokenizer_file_path)
+    
+    def save_tokenizer_config(self, save_dir: Path) -> None:
+        # convert Path to str
+        for k in self.tokenizer_config:
+            if isinstance(self.tokenizer_config[k], Path):
+                self.tokenizer_config[k] = str(self.tokenizer_config[k])
+
+        info_file = save_dir / "tokenizer_config.json"
+        with info_file.open("w") as f:
+            json.dump(self.tokenizer_config, f, indent=4)
+            
+    def load_json(self, path: Path) -> dict:
+        with path.open("r") as f:
+            return json.load(f)
+        
+class SPTokenizer(HFGPTXTokenizer):
+    model_file_glob = "*tokenizer.model"
+    vocab_files_names = {"tokenizer_file": "tokenizer.model"}
+    decode_kwargs = ["num_threads"]
+    # `is_continuation` does not work without this, but it doesn't
+    # implement all APIs of `PreTrainedTokenizer`.
+    def encode(self, text: str, **kwargs) -> List[int]:
+        return_tokens = kwargs.pop('return_tokens', False)
+        is_continuation = kwargs.pop('is_continuation', False)
+        return self._encode(
+            text,
+            return_tokens=return_tokens,
+            is_continuation=is_continuation,
+        )
+        
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+
+        self.eos_token = "</s>"
+        self.eos_token_id = 2
+        self.system_messages_by_lang = {  # translations by deepl / google translate
+                "BG": "Чат между човек и асистент с изкуствен интелект. Асистентът дава полезни и учтиви отговори на въпросите на човека.",  # noqa
+                "CS": "Chat mezi člověkem a asistentem s umělou inteligencí. Asistent poskytuje vstřícné a zdvořilé odpovědi na otázky člověka.",  # noqa
+                "DA": "En chat mellem et menneske og en assistent med kunstig intelligens, som giver hjælpsomme og høflige svar på menneskets spørgsmål.",  # noqa
+                "DE": "Ein Gespräch zwischen einem Menschen und einem Assistenten mit künstlicher Intelligenz. Der Assistent gibt hilfreiche und höfliche Antworten auf die Fragen des Menschen.",  # noqa
+                "EL": "Μια συνομιλία μεταξύ ενός ανθρώπου και ενός βοηθού τεχνητής νοημοσύνης. Ο βοηθός δίνει χρήσιμες και ευγενικές απαντήσεις στις ερωτήσεις του ανθρώπου.",  # noqa
+                "EN": "A chat between a human and an artificial intelligence assistant.The assistant gives helpful and polite answers to the human's questions.",  # noqa
+                "ES": "Una conversación entre un humano y un asistente de inteligencia artificial. El asistente da respuestas útiles y amables a las preguntas del humano.",  # noqa
+                "ET": "Inimese ja tehisintellekti assistendi vaheline vestlus. Assistent annab inimese küsimustele abivalmis ja viisakaid vastuseid.",  # noqa
+                "FI": "Ihmisen ja tekoälyavustajan välinen keskustelu. Avustaja antaa avuliaita ja kohteliaita vastauksia ihmisen kysymyksiin.",  # noqa
+                "FR": "Conversation entre un humain et un assistant doté d'une intelligence artificielle. L'assistant donne des réponses utiles et polies aux questions de l'homme.",  # noqa
+                "GA": "Comhrá idir duine agus cúntóir hintleachta saorga. Tugann an cúntóir freagraí cabhracha dea-bhéasacha ar cheisteanna an duine.",  # noqa
+                "HR": "Razgovor između čovjeka i pomoćnika umjetne inteligencije. Pomoćnik daje korisne i ljubazne odgovore na ljudska pitanja.",  # noqa
+                "HU": "Egy ember és egy mesterséges intelligencia asszisztens közötti beszélgetés. Az asszisztens segítőkész és udvarias válaszokat ad az ember kérdéseire.",  # noqa
+                "IT": "Una chat tra un umano e un assistente di intelligenza artificiale. L'assistente fornisce risposte utili ed educate alle domande dell'uomo.",  # noqa
+                "LT": "Žmogaus ir dirbtinio intelekto asistento pokalbis. Asistentas naudingai ir mandagiai atsako į žmogaus klausimus.",  # noqa
+                "LV": "Cilvēka un mākslīgā intelekta asistenta tērzēšana. Asistents sniedz noderīgas un pieklājīgas atbildes uz cilvēka jautājumiem.",  # noqa
+                "MT": "Chat bejn bniedem u assistent ta' intelliġenza artifiċjali. L-assistent jagħti tweġibiet ta' għajnuna u edukat għall-mistoqsijiet tal-bniedem.",  # noqa
+                "NL": "Een chat tussen een mens en een assistent met kunstmatige intelligentie. De assistent geeft behulpzame en beleefde antwoorden op de vragen van de mens.",  # noqa
+                "PL": "Czat między człowiekiem a asystentem sztucznej inteligencji. Asystent udziela pomocnych i uprzejmych odpowiedzi na pytania człowieka.",  # noqa
+                "PT": "Uma conversa entre um ser humano e um assistente de inteligência artificial. O assistente dá respostas úteis e educadas às perguntas do utilizador.",  # noqa
+                "RO": "O conversație între un om și un asistent cu inteligență artificială. Asistentul oferă răspunsuri utile și politicoase la întrebările omului.",  # noqa
+                "SK": "Rozhovor medzi človekom a asistentom s umelou inteligenciou. Asistent poskytuje užitočné a zdvorilé odpovede na otázky človeka.",  # noqa
+                "SL": "Pogovor med človekom in pomočnikom z umetno inteligenco. Pomočnik človeku prijazno in vljudno odgovarja na njegova vprašanja.",  # noqa
+                "SV": "En chatt mellan en människa och en assistent med artificiell intelligens. Assistenten ger hjälpsamma och artiga svar på människans frågor.",  # noqa
+        }
+        chat_template = "{%- for message in messages %}\n{%- if (message['role']|lower == 'user') != (loop.index0 % 2 == 0) %}\n{{- raise_exception('Roles must alternate User/Assistant/User/Assistant/...') }}\n{%- endif %}\n{%-if message['role']|lower == 'user' %}\n{{- message['role']|capitalize + ': ' + message['content'] + '\\n' }}\n{%- elif message['role']|lower == 'assistant' %}\n{{- message['role']|capitalize + ': ' + message['content'] + eos_token + '\\n' }}\n{%- else %}\n{{- raise_exception('Only user and assistant roles are supported!') }}\n {%- endif %}\n{%- endfor %}{%-if add_generation_prompt %}\n{{- 'Assistant: '}}\n{%- endif %}\n"
+        self.chat_template = {
+            lang: f"System: {sys_msg}" + "{{- '\\n'}}\n" + chat_template
+            for lang, sys_msg in self.system_messages_by_lang.items()
+        }
+        self.chat_template['default'] = f"System: {self.system_messages_by_lang['EN']}" + "{{- '\\n'}}\n" + chat_template

special_tokens_map.json

@@ -0,0 +1,281 @@
+{
+  "additional_special_tokens": [
+    "<s>",
+    "</s>",
+    "<pad>",
+    "<eod>",
+    "<placeholder_tok_0>",
+    "<placeholder_tok_1>",
+    "<placeholder_tok_2>",
+    "<placeholder_tok_3>",
+    "<placeholder_tok_4>",
+    "<placeholder_tok_5>",
+    "<placeholder_tok_6>",
+    "<placeholder_tok_7>",
+    "<placeholder_tok_8>",
+    "<placeholder_tok_9>",
+    "<placeholder_tok_10>",
+    "<placeholder_tok_11>",
+    "<placeholder_tok_12>",
+    "<placeholder_tok_13>",
+    "<placeholder_tok_14>",
+    "<placeholder_tok_15>",
+    "<placeholder_tok_16>",
+    "<placeholder_tok_17>",
+    "<placeholder_tok_18>",
+    "<placeholder_tok_19>",
+    "<placeholder_tok_20>",
+    "<placeholder_tok_21>",
+    "<placeholder_tok_22>",
+    "<placeholder_tok_23>",
+    "<placeholder_tok_24>",
+    "<placeholder_tok_25>",
+    "<placeholder_tok_26>",
+    "<placeholder_tok_27>",
+    "<placeholder_tok_28>",
+    "<placeholder_tok_29>",
+    "<placeholder_tok_30>",
+    "<placeholder_tok_31>",
+    "<placeholder_tok_32>",
+    "<placeholder_tok_33>",
+    "<placeholder_tok_34>",
+    "<placeholder_tok_35>",
+    "<placeholder_tok_36>",
+    "<placeholder_tok_37>",
+    "<placeholder_tok_38>",
+    "<placeholder_tok_39>",
+    "<placeholder_tok_40>",
+    "<placeholder_tok_41>",
+    "<placeholder_tok_42>",
+    "<placeholder_tok_43>",
+    "<placeholder_tok_44>",
+    "<placeholder_tok_45>",
+    "<placeholder_tok_46>",
+    "<placeholder_tok_47>",
+    "<placeholder_tok_48>",
+    "<placeholder_tok_49>",
+    "<placeholder_tok_50>",
+    "<placeholder_tok_51>",
+    "<placeholder_tok_52>",
+    "<placeholder_tok_53>",
+    "<placeholder_tok_54>",
+    "<placeholder_tok_55>",
+    "<placeholder_tok_56>",
+    "<placeholder_tok_57>",
+    "<placeholder_tok_58>",
+    "<placeholder_tok_59>",
+    "<placeholder_tok_60>",
+    "<placeholder_tok_61>",
+    "<placeholder_tok_62>",
+    "<placeholder_tok_63>",
+    "<placeholder_tok_64>",
+    "<placeholder_tok_65>",
+    "<placeholder_tok_66>",
+    "<placeholder_tok_67>",
+    "<placeholder_tok_68>",
+    "<placeholder_tok_69>",
+    "<placeholder_tok_70>",
+    "<placeholder_tok_71>",
+    "<placeholder_tok_72>",
+    "<placeholder_tok_73>",
+    "<placeholder_tok_74>",
+    "<placeholder_tok_75>",
+    "<placeholder_tok_76>",
+    "<placeholder_tok_77>",
+    "<placeholder_tok_78>",
+    "<placeholder_tok_79>",
+    "<placeholder_tok_80>",
+    "<placeholder_tok_81>",
+    "<placeholder_tok_82>",
+    "<placeholder_tok_83>",
+    "<placeholder_tok_84>",
+    "<placeholder_tok_85>",
+    "<placeholder_tok_86>",
+    "<placeholder_tok_87>",
+    "<placeholder_tok_88>",
+    "<placeholder_tok_89>",
+    "<placeholder_tok_90>",
+    "<placeholder_tok_91>",
+    "<placeholder_tok_92>",
+    "<placeholder_tok_93>",
+    "<placeholder_tok_94>",
+    "<placeholder_tok_95>",
+    "<placeholder_tok_96>",
+    "<placeholder_tok_97>",
+    "<placeholder_tok_98>",
+    "<placeholder_tok_99>",
+    "<placeholder_tok_100>",
+    "<placeholder_tok_101>",
+    "<placeholder_tok_102>",
+    "<placeholder_tok_103>",
+    "<placeholder_tok_104>",
+    "<placeholder_tok_105>",
+    "<placeholder_tok_106>",
+    "<placeholder_tok_107>",
+    "<placeholder_tok_108>",
+    "<placeholder_tok_109>",
+    "<placeholder_tok_110>",
+    "<placeholder_tok_111>",
+    "<placeholder_tok_112>",
+    "<placeholder_tok_113>",
+    "<placeholder_tok_114>",
+    "<placeholder_tok_115>",
+    "<placeholder_tok_116>",
+    "<placeholder_tok_117>",
+    "<placeholder_tok_118>",
+    "<placeholder_tok_119>",
+    "<placeholder_tok_120>",
+    "<placeholder_tok_121>",
+    "<placeholder_tok_122>",
+    "<placeholder_tok_123>",
+    "<placeholder_tok_124>",
+    "<placeholder_tok_125>",
+    "<placeholder_tok_126>",
+    "<placeholder_tok_127>",
+    "<placeholder_tok_128>",
+    "<placeholder_tok_129>",
+    "<placeholder_tok_130>",
+    "<placeholder_tok_131>",
+    "<placeholder_tok_132>",
+    "<placeholder_tok_133>",
+    "<placeholder_tok_134>",
+    "<placeholder_tok_135>",
+    "<placeholder_tok_136>",
+    "<placeholder_tok_137>",
+    "<placeholder_tok_138>",
+    "<placeholder_tok_139>",
+    "<placeholder_tok_140>",
+    "<placeholder_tok_141>",
+    "<placeholder_tok_142>",
+    "<placeholder_tok_143>",
+    "<placeholder_tok_144>",
+    "<placeholder_tok_145>",
+    "<placeholder_tok_146>",
+    "<placeholder_tok_147>",
+    "<placeholder_tok_148>",
+    "<placeholder_tok_149>",
+    "<placeholder_tok_150>",
+    "<placeholder_tok_151>",
+    "<placeholder_tok_152>",
+    "<placeholder_tok_153>",
+    "<placeholder_tok_154>",
+    "<placeholder_tok_155>",
+    "<placeholder_tok_156>",
+    "<placeholder_tok_157>",
+    "<placeholder_tok_158>",
+    "<placeholder_tok_159>",
+    "<placeholder_tok_160>",
+    "<placeholder_tok_161>",
+    "<placeholder_tok_162>",
+    "<placeholder_tok_163>",
+    "<placeholder_tok_164>",
+    "<placeholder_tok_165>",
+    "<placeholder_tok_166>",
+    "<placeholder_tok_167>",
+    "<placeholder_tok_168>",
+    "<placeholder_tok_169>",
+    "<placeholder_tok_170>",
+    "<placeholder_tok_171>",
+    "<placeholder_tok_172>",
+    "<placeholder_tok_173>",
+    "<placeholder_tok_174>",
+    "<placeholder_tok_175>",
+    "<placeholder_tok_176>",
+    "<placeholder_tok_177>",
+    "<placeholder_tok_178>",
+    "<placeholder_tok_179>",
+    "<placeholder_tok_180>",
+    "<placeholder_tok_181>",
+    "<placeholder_tok_182>",
+    "<placeholder_tok_183>",
+    "<placeholder_tok_184>",
+    "<placeholder_tok_185>",
+    "<placeholder_tok_186>",
+    "<placeholder_tok_187>",
+    "<placeholder_tok_188>",
+    "<placeholder_tok_189>",
+    "<placeholder_tok_190>",
+    "<placeholder_tok_191>",
+    "<placeholder_tok_192>",
+    "<placeholder_tok_193>",
+    "<placeholder_tok_194>",
+    "<placeholder_tok_195>",
+    "<placeholder_tok_196>",
+    "<placeholder_tok_197>",
+    "<placeholder_tok_198>",
+    "<placeholder_tok_199>",
+    "<placeholder_tok_200>",
+    "<placeholder_tok_201>",
+    "<placeholder_tok_202>",
+    "<placeholder_tok_203>",
+    "<placeholder_tok_204>",
+    "<placeholder_tok_205>",
+    "<placeholder_tok_206>",
+    "<placeholder_tok_207>",
+    "<placeholder_tok_208>",
+    "<placeholder_tok_209>",
+    "<placeholder_tok_210>",
+    "<placeholder_tok_211>",
+    "<placeholder_tok_212>",
+    "<placeholder_tok_213>",
+    "<placeholder_tok_214>",
+    "<placeholder_tok_215>",
+    "<placeholder_tok_216>",
+    "<placeholder_tok_217>",
+    "<placeholder_tok_218>",
+    "<placeholder_tok_219>",
+    "<placeholder_tok_220>",
+    "<placeholder_tok_221>",
+    "<placeholder_tok_222>",
+    "<placeholder_tok_223>",
+    "<placeholder_tok_224>",
+    "<placeholder_tok_225>",
+    "<placeholder_tok_226>",
+    "<placeholder_tok_227>",
+    "<placeholder_tok_228>",
+    "<placeholder_tok_229>",
+    "<placeholder_tok_230>",
+    "<placeholder_tok_231>",
+    "<placeholder_tok_232>",
+    "<placeholder_tok_233>",
+    "<placeholder_tok_234>",
+    "<placeholder_tok_235>",
+    "<placeholder_tok_236>",
+    "<placeholder_tok_237>",
+    "<placeholder_tok_238>",
+    "<placeholder_tok_239>",
+    "<placeholder_tok_240>",
+    "<placeholder_tok_241>",
+    "<placeholder_tok_242>",
+    "<placeholder_tok_243>",
+    "<placeholder_tok_244>",
+    "<placeholder_tok_245>",
+    "<placeholder_tok_246>",
+    "<placeholder_tok_247>",
+    "<placeholder_tok_248>",
+    "<placeholder_tok_249>",
+    "<placeholder_tok_250>",
+    "<placeholder_tok_251>",
+    "<placeholder_tok_252>",
+    "<placeholder_tok_253>",
+    "<placeholder_tok_254>",
+    "<placeholder_tok_255>",
+    {
+      "content": "<|im_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false
+    }
+  ],
+  "bos_token": "<s>",
+  "eos_token": {
+    "content": "<|im_end|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": "<pad>",
+  "unk_token": "<unk>"
+}

tokenizer.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:08d0c8316539a853f2fe6e14f51f0df583011dfb078fa08c8b6dc5c15a19a7e6
+size 4719922

tokenizer_config.json

@@ -0,0 +1,292 @@
+{
+    "num_threads": 224,
+    "split_by_whitespace": true,
+    "model_type": "unigram",
+    "vocab_size": 250680,
+    "character_coverage": 0.9999,
+    "byte_fallback": true,
+    "split_by_number": true,
+    "split_digits": true,
+    "normalization_rule_name": "nfkc",
+    "max_sentence_length": 4096,
+    "shuffle_input_sentence": true,
+    "input_sentence_size": 0,
+    "train_extremely_large_corpus": true,
+    "allow_whitespace_only_pieces": true,
+    "required_chars": "",
+    "remove_extra_whitespaces": false,
+    "user_defined_symbols": [
+        "<s>",
+        "</s>",
+        "<pad>",
+        "<eod>",
+        "<placeholder_tok_0>",
+        "<placeholder_tok_1>",
+        "<placeholder_tok_2>",
+        "<placeholder_tok_3>",
+        "<placeholder_tok_4>",
+        "<placeholder_tok_5>",
+        "<placeholder_tok_6>",
+        "<placeholder_tok_7>",
+        "<placeholder_tok_8>",
+        "<placeholder_tok_9>",
+        "<placeholder_tok_10>",
+        "<placeholder_tok_11>",
+        "<placeholder_tok_12>",
+        "<placeholder_tok_13>",
+        "<placeholder_tok_14>",
+        "<placeholder_tok_15>",
+        "<placeholder_tok_16>",
+        "<placeholder_tok_17>",
+        "<placeholder_tok_18>",
+        "<placeholder_tok_19>",
+        "<placeholder_tok_20>",
+        "<placeholder_tok_21>",
+        "<placeholder_tok_22>",
+        "<placeholder_tok_23>",
+        "<placeholder_tok_24>",
+        "<placeholder_tok_25>",
+        "<placeholder_tok_26>",
+        "<placeholder_tok_27>",
+        "<placeholder_tok_28>",
+        "<placeholder_tok_29>",
+        "<placeholder_tok_30>",
+        "<placeholder_tok_31>",
+        "<placeholder_tok_32>",
+        "<placeholder_tok_33>",
+        "<placeholder_tok_34>",
+        "<placeholder_tok_35>",
+        "<placeholder_tok_36>",
+        "<placeholder_tok_37>",
+        "<placeholder_tok_38>",
+        "<placeholder_tok_39>",
+        "<placeholder_tok_40>",
+        "<placeholder_tok_41>",
+        "<placeholder_tok_42>",
+        "<placeholder_tok_43>",
+        "<placeholder_tok_44>",
+        "<placeholder_tok_45>",
+        "<placeholder_tok_46>",
+        "<placeholder_tok_47>",
+        "<placeholder_tok_48>",
+        "<placeholder_tok_49>",
+        "<placeholder_tok_50>",
+        "<placeholder_tok_51>",
+        "<placeholder_tok_52>",
+        "<placeholder_tok_53>",
+        "<placeholder_tok_54>",
+        "<placeholder_tok_55>",
+        "<placeholder_tok_56>",
+        "<placeholder_tok_57>",
+        "<placeholder_tok_58>",
+        "<placeholder_tok_59>",
+        "<placeholder_tok_60>",
+        "<placeholder_tok_61>",
+        "<placeholder_tok_62>",
+        "<placeholder_tok_63>",
+        "<placeholder_tok_64>",
+        "<placeholder_tok_65>",
+        "<placeholder_tok_66>",
+        "<placeholder_tok_67>",
+        "<placeholder_tok_68>",
+        "<placeholder_tok_69>",
+        "<placeholder_tok_70>",
+        "<placeholder_tok_71>",
+        "<placeholder_tok_72>",
+        "<placeholder_tok_73>",
+        "<placeholder_tok_74>",
+        "<placeholder_tok_75>",
+        "<placeholder_tok_76>",
+        "<placeholder_tok_77>",
+        "<placeholder_tok_78>",
+        "<placeholder_tok_79>",
+        "<placeholder_tok_80>",
+        "<placeholder_tok_81>",
+        "<placeholder_tok_82>",
+        "<placeholder_tok_83>",
+        "<placeholder_tok_84>",
+        "<placeholder_tok_85>",
+        "<placeholder_tok_86>",
+        "<placeholder_tok_87>",
+        "<placeholder_tok_88>",
+        "<placeholder_tok_89>",
+        "<placeholder_tok_90>",
+        "<placeholder_tok_91>",
+        "<placeholder_tok_92>",
+        "<placeholder_tok_93>",
+        "<placeholder_tok_94>",
+        "<placeholder_tok_95>",
+        "<placeholder_tok_96>",
+        "<placeholder_tok_97>",
+        "<placeholder_tok_98>",
+        "<placeholder_tok_99>",
+        "<placeholder_tok_100>",
+        "<placeholder_tok_101>",
+        "<placeholder_tok_102>",
+        "<placeholder_tok_103>",
+        "<placeholder_tok_104>",
+        "<placeholder_tok_105>",
+        "<placeholder_tok_106>",
+        "<placeholder_tok_107>",
+        "<placeholder_tok_108>",
+        "<placeholder_tok_109>",
+        "<placeholder_tok_110>",
+        "<placeholder_tok_111>",
+        "<placeholder_tok_112>",
+        "<placeholder_tok_113>",
+        "<placeholder_tok_114>",
+        "<placeholder_tok_115>",
+        "<placeholder_tok_116>",
+        "<placeholder_tok_117>",
+        "<placeholder_tok_118>",
+        "<placeholder_tok_119>",
+        "<placeholder_tok_120>",
+        "<placeholder_tok_121>",
+        "<placeholder_tok_122>",
+        "<placeholder_tok_123>",
+        "<placeholder_tok_124>",
+        "<placeholder_tok_125>",
+        "<placeholder_tok_126>",
+        "<placeholder_tok_127>",
+        "<placeholder_tok_128>",
+        "<placeholder_tok_129>",
+        "<placeholder_tok_130>",
+        "<placeholder_tok_131>",
+        "<placeholder_tok_132>",
+        "<placeholder_tok_133>",
+        "<placeholder_tok_134>",
+        "<placeholder_tok_135>",
+        "<placeholder_tok_136>",
+        "<placeholder_tok_137>",
+        "<placeholder_tok_138>",
+        "<placeholder_tok_139>",
+        "<placeholder_tok_140>",
+        "<placeholder_tok_141>",
+        "<placeholder_tok_142>",
+        "<placeholder_tok_143>",
+        "<placeholder_tok_144>",
+        "<placeholder_tok_145>",
+        "<placeholder_tok_146>",
+        "<placeholder_tok_147>",
+        "<placeholder_tok_148>",
+        "<placeholder_tok_149>",
+        "<placeholder_tok_150>",
+        "<placeholder_tok_151>",
+        "<placeholder_tok_152>",
+        "<placeholder_tok_153>",
+        "<placeholder_tok_154>",
+        "<placeholder_tok_155>",
+        "<placeholder_tok_156>",
+        "<placeholder_tok_157>",
+        "<placeholder_tok_158>",
+        "<placeholder_tok_159>",
+        "<placeholder_tok_160>",
+        "<placeholder_tok_161>",
+        "<placeholder_tok_162>",
+        "<placeholder_tok_163>",
+        "<placeholder_tok_164>",
+        "<placeholder_tok_165>",
+        "<placeholder_tok_166>",
+        "<placeholder_tok_167>",
+        "<placeholder_tok_168>",
+        "<placeholder_tok_169>",
+        "<placeholder_tok_170>",
+        "<placeholder_tok_171>",
+        "<placeholder_tok_172>",
+        "<placeholder_tok_173>",
+        "<placeholder_tok_174>",
+        "<placeholder_tok_175>",
+        "<placeholder_tok_176>",
+        "<placeholder_tok_177>",
+        "<placeholder_tok_178>",
+        "<placeholder_tok_179>",
+        "<placeholder_tok_180>",
+        "<placeholder_tok_181>",
+        "<placeholder_tok_182>",
+        "<placeholder_tok_183>",
+        "<placeholder_tok_184>",
+        "<placeholder_tok_185>",
+        "<placeholder_tok_186>",
+        "<placeholder_tok_187>",
+        "<placeholder_tok_188>",
+        "<placeholder_tok_189>",
+        "<placeholder_tok_190>",
+        "<placeholder_tok_191>",
+        "<placeholder_tok_192>",
+        "<placeholder_tok_193>",
+        "<placeholder_tok_194>",
+        "<placeholder_tok_195>",
+        "<placeholder_tok_196>",
+        "<placeholder_tok_197>",
+        "<placeholder_tok_198>",
+        "<placeholder_tok_199>",
+        "<placeholder_tok_200>",
+        "<placeholder_tok_201>",
+        "<placeholder_tok_202>",
+        "<placeholder_tok_203>",
+        "<placeholder_tok_204>",
+        "<placeholder_tok_205>",
+        "<placeholder_tok_206>",
+        "<placeholder_tok_207>",
+        "<placeholder_tok_208>",
+        "<placeholder_tok_209>",
+        "<placeholder_tok_210>",
+        "<placeholder_tok_211>",
+        "<placeholder_tok_212>",
+        "<placeholder_tok_213>",
+        "<placeholder_tok_214>",
+        "<placeholder_tok_215>",
+        "<placeholder_tok_216>",
+        "<placeholder_tok_217>",
+        "<placeholder_tok_218>",
+        "<placeholder_tok_219>",
+        "<placeholder_tok_220>",
+        "<placeholder_tok_221>",
+        "<placeholder_tok_222>",
+        "<placeholder_tok_223>",
+        "<placeholder_tok_224>",
+        "<placeholder_tok_225>",
+        "<placeholder_tok_226>",
+        "<placeholder_tok_227>",
+        "<placeholder_tok_228>",
+        "<placeholder_tok_229>",
+        "<placeholder_tok_230>",
+        "<placeholder_tok_231>",
+        "<placeholder_tok_232>",
+        "<placeholder_tok_233>",
+        "<placeholder_tok_234>",
+        "<placeholder_tok_235>",
+        "<placeholder_tok_236>",
+        "<placeholder_tok_237>",
+        "<placeholder_tok_238>",
+        "<placeholder_tok_239>",
+        "<placeholder_tok_240>",
+        "<placeholder_tok_241>",
+        "<placeholder_tok_242>",
+        "<placeholder_tok_243>",
+        "<placeholder_tok_244>",
+        "<placeholder_tok_245>",
+        "<placeholder_tok_246>",
+        "<placeholder_tok_247>",
+        "<placeholder_tok_248>",
+        "<placeholder_tok_249>",
+        "<placeholder_tok_250>",
+        "<placeholder_tok_251>",
+        "<placeholder_tok_252>",
+        "<placeholder_tok_253>",
+        "<placeholder_tok_254>",
+        "<placeholder_tok_255>"
+    ],
+    "datasets_dir": "/home/fhgiais/gptx_ablations/bias_analysis/data/tokenizer/temp/",
+    "save_dir": "/home/fhgiais/gptx_ablations/bias_analysis/tokenizer/24",
+    "text_key": "text",
+    "cache_dir": "/home/fhgiais/gptx_ablations/bias_analysis/tokenizer/24/cache",
+    "library": "sentencepiece",
+    "auto_map": {
+        "AutoTokenizer": [
+            "gptx_tokenizer.SPTokenizer",
+            null
+        ]
+    },
+    "tokenizer_class": "SPTokenizer"
+}