grok_tokenizer

grok_tokenizer#

Bases: BaseTokenizer “)

Source Code in griptape/tokenizers/grok_tokenizer.py

@define()
class GrokTokenizer(BaseTokenizer):
    # https://docs.x.ai/docs/models?cluster=us-east-1#model-constraints
    MODEL_PREFIXES_TO_MAX_INPUT_TOKENS = {
        "grok-2-vision": 32768,
        "grok-2": 131072,
        "grok-vision-beta": 8192,
        "grok-beta": 131072,
    }

    MODEL_PREFIXES_TO_MAX_OUTPUT_TOKENS = {"grok": 4096}

    base_url: str = field(default="https://api.x.ai", kw_only=True, metadata={"serializable": True})
    api_key: str = field(kw_only=True, default=None)
    headers: dict = field(
        default=Factory(lambda self: {"Authorization": f"Bearer {self.api_key}"}, takes_self=True), kw_only=True
    )

    def count_tokens(self, text: str) -> int:
        response = requests.post(
            urljoin(self.base_url, "/v1/tokenize-text"),
            headers=self.headers,
            json={"text": text, "model": self.model},
        )
        response.raise_for_status()
        return len(response.json()["token_ids"])

  • MODEL_PREFIXES_TO_MAX_INPUT_TOKENS = {'grok-2-vision': 32768, 'grok-2': 131072, 'grok-vision-beta': 8192, 'grok-beta': 131072} class-attribute instance-attribute

  • MODEL_PREFIXES_TO_MAX_OUTPUT_TOKENS = {'grok': 4096} class-attribute instance-attribute

  • api_key = field(kw_only=True, default=None) class-attribute instance-attribute

  • base_url = field(default='https://api.x.ai', kw_only=True, metadata={'serializable': True}) class-attribute instance-attribute

  • headers = field(default=Factory(lambda self: {'Authorization': f'Bearer {self.api_key}'}, takes_self=True), kw_only=True) class-attribute instance-attribute

count_tokens(text)#

Source Code in griptape/tokenizers/grok_tokenizer.py

def count_tokens(self, text: str) -> int:
    response = requests.post(
        urljoin(self.base_url, "/v1/tokenize-text"),
        headers=self.headers,
        json={"text": text, "model": self.model},
    )
    response.raise_for_status()
    return len(response.json()["token_ids"])
Contents