Skip to content

Openai tokenizer

OpenAiTokenizer

Bases: BaseTokenizer

Source code in griptape/tokenizers/openai_tokenizer.py 
  9
 10
 11
 12
 13
 14
 15
 16
 17
 18
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
@define()
class OpenAiTokenizer(BaseTokenizer):
    DEFAULT_OPENAI_GPT_3_COMPLETION_MODEL = "gpt-3.5-turbo-instruct"
    DEFAULT_OPENAI_GPT_3_CHAT_MODEL = "gpt-3.5-turbo"
    DEFAULT_OPENAI_GPT_4_MODEL = "gpt-4"
    DEFAULT_ENCODING = "cl100k_base"
    DEFAULT_MAX_TOKENS = 2049
    DEFAULT_MAX_OUTPUT_TOKENS = 4096
    TOKEN_OFFSET = 8

    # https://platform.openai.com/docs/models/gpt-4-and-gpt-4-turbo
    MODEL_PREFIXES_TO_MAX_INPUT_TOKENS = {
        "gpt-4-1106": 128000,
        "gpt-4-32k": 32768,
        "gpt-4": 8192,
        "gpt-3.5-turbo-16k": 16384,
        "gpt-3.5-turbo": 4096,
        "gpt-35-turbo-16k": 16384,
        "gpt-35-turbo": 4096,
        "text-embedding-ada-002": 8191,
        "text-embedding-ada-001": 2046,
        "text-embedding-3-small": 8191,
        "text-embedding-3-large": 8191,
    }

    MODEL_PREFIXES_TO_MAX_OUTPUT_TOKENS = {"gpt": 4096}

    EMBEDDING_MODELS = [
        "text-embedding-ada-002",
        "text-embedding-ada-001",
        "text-embedding-3-small",
        "text-embedding-3-large",
    ]

    @property
    def encoding(self) -> tiktoken.Encoding:
        try:
            return tiktoken.encoding_for_model(self.model)
        except KeyError:
            return tiktoken.get_encoding(self.DEFAULT_ENCODING)

    def _default_max_input_tokens(self) -> int:
        tokens = next((v for k, v in self.MODEL_PREFIXES_TO_MAX_INPUT_TOKENS.items() if self.model.startswith(k)), None)
        offset = 0 if self.model in self.EMBEDDING_MODELS else self.TOKEN_OFFSET

        return (tokens if tokens else self.DEFAULT_MAX_TOKENS) - offset

    def _default_max_output_tokens(self) -> int:
        tokens = next(
            (v for k, v in self.MODEL_PREFIXES_TO_MAX_OUTPUT_TOKENS.items() if self.model.startswith(k)), None
        )

        if tokens is None:
            return self.DEFAULT_MAX_OUTPUT_TOKENS
        else:
            return tokens

    def count_tokens(self, text: str | list[dict], model: Optional[str] = None) -> int:
        """
        Handles the special case of ChatML. Implementation adopted from the official OpenAI notebook:
        https://github.com/openai/openai-cookbook/blob/main/examples/How_to_count_tokens_with_tiktoken.ipynb
        """
        if isinstance(text, list):
            model = model if model else self.model

            try:
                encoding = tiktoken.encoding_for_model(model)
            except KeyError:
                logging.warning("model not found. Using cl100k_base encoding.")

                encoding = tiktoken.get_encoding("cl100k_base")

            if model in {
                "gpt-3.5-turbo-0613",
                "gpt-3.5-turbo-16k-0613",
                "gpt-4-0314",
                "gpt-4-32k-0314",
                "gpt-4-0613",
                "gpt-4-32k-0613",
            }:
                tokens_per_message = 3
                tokens_per_name = 1
            elif model == "gpt-3.5-turbo-0301":
                # every message follows <|start|>{role/name}\n{content}<|end|>\n
                tokens_per_message = 4
                # if there's a name, the role is omitted
                tokens_per_name = -1
            elif "gpt-3.5-turbo" in model or "gpt-35-turbo" in model:
                logging.info("gpt-3.5-turbo may update over time. Returning num tokens assuming gpt-3.5-turbo-0613.")
                return self.count_tokens(text, model="gpt-3.5-turbo-0613")
            elif "gpt-4" in model:
                logging.info("gpt-4 may update over time. Returning num tokens assuming gpt-4-0613.")
                return self.count_tokens(text, model="gpt-4-0613")
            else:
                raise NotImplementedError(
                    f"""token_count() is not implemented for model {model}. 
                    See https://github.com/openai/openai-python/blob/main/chatml.md for 
                    information on how messages are converted to tokens."""
                )

            num_tokens = 0

            for message in text:
                num_tokens += tokens_per_message
                for key, value in message.items():
                    num_tokens += len(encoding.encode(value))
                    if key == "name":
                        num_tokens += tokens_per_name

            # every reply is primed with <|start|>assistant<|message|>
            num_tokens += 3

            return num_tokens
        else:
            return len(self.encoding.encode(text, allowed_special=set(self.stop_sequences)))

DEFAULT_ENCODING = 'cl100k_base' class-attribute instance-attribute

DEFAULT_MAX_OUTPUT_TOKENS = 4096 class-attribute instance-attribute

DEFAULT_MAX_TOKENS = 2049 class-attribute instance-attribute

DEFAULT_OPENAI_GPT_3_CHAT_MODEL = 'gpt-3.5-turbo' class-attribute instance-attribute

DEFAULT_OPENAI_GPT_3_COMPLETION_MODEL = 'gpt-3.5-turbo-instruct' class-attribute instance-attribute

DEFAULT_OPENAI_GPT_4_MODEL = 'gpt-4' class-attribute instance-attribute

EMBEDDING_MODELS = ['text-embedding-ada-002', 'text-embedding-ada-001', 'text-embedding-3-small', 'text-embedding-3-large'] class-attribute instance-attribute

MODEL_PREFIXES_TO_MAX_INPUT_TOKENS = {'gpt-4-1106': 128000, 'gpt-4-32k': 32768, 'gpt-4': 8192, 'gpt-3.5-turbo-16k': 16384, 'gpt-3.5-turbo': 4096, 'gpt-35-turbo-16k': 16384, 'gpt-35-turbo': 4096, 'text-embedding-ada-002': 8191, 'text-embedding-ada-001': 2046, 'text-embedding-3-small': 8191, 'text-embedding-3-large': 8191} class-attribute instance-attribute

MODEL_PREFIXES_TO_MAX_OUTPUT_TOKENS = {'gpt': 4096} class-attribute instance-attribute

TOKEN_OFFSET = 8 class-attribute instance-attribute

encoding: tiktoken.Encoding property

count_tokens(text, model=None)

Handles the special case of ChatML. Implementation adopted from the official OpenAI notebook: https://github.com/openai/openai-cookbook/blob/main/examples/How_to_count_tokens_with_tiktoken.ipynb

Source code in griptape/tokenizers/openai_tokenizer.py 
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
def count_tokens(self, text: str | list[dict], model: Optional[str] = None) -> int:
    """
    Handles the special case of ChatML. Implementation adopted from the official OpenAI notebook:
    https://github.com/openai/openai-cookbook/blob/main/examples/How_to_count_tokens_with_tiktoken.ipynb
    """
    if isinstance(text, list):
        model = model if model else self.model

        try:
            encoding = tiktoken.encoding_for_model(model)
        except KeyError:
            logging.warning("model not found. Using cl100k_base encoding.")

            encoding = tiktoken.get_encoding("cl100k_base")

        if model in {
            "gpt-3.5-turbo-0613",
            "gpt-3.5-turbo-16k-0613",
            "gpt-4-0314",
            "gpt-4-32k-0314",
            "gpt-4-0613",
            "gpt-4-32k-0613",
        }:
            tokens_per_message = 3
            tokens_per_name = 1
        elif model == "gpt-3.5-turbo-0301":
            # every message follows <|start|>{role/name}\n{content}<|end|>\n
            tokens_per_message = 4
            # if there's a name, the role is omitted
            tokens_per_name = -1
        elif "gpt-3.5-turbo" in model or "gpt-35-turbo" in model:
            logging.info("gpt-3.5-turbo may update over time. Returning num tokens assuming gpt-3.5-turbo-0613.")
            return self.count_tokens(text, model="gpt-3.5-turbo-0613")
        elif "gpt-4" in model:
            logging.info("gpt-4 may update over time. Returning num tokens assuming gpt-4-0613.")
            return self.count_tokens(text, model="gpt-4-0613")
        else:
            raise NotImplementedError(
                f"""token_count() is not implemented for model {model}. 
                See https://github.com/openai/openai-python/blob/main/chatml.md for 
                information on how messages are converted to tokens."""
            )

        num_tokens = 0

        for message in text:
            num_tokens += tokens_per_message
            for key, value in message.items():
                num_tokens += len(encoding.encode(value))
                if key == "name":
                    num_tokens += tokens_per_name

        # every reply is primed with <|start|>assistant<|message|>
        num_tokens += 3

        return num_tokens
    else:
        return len(self.encoding.encode(text, allowed_special=set(self.stop_sequences)))