def __init__()

in src/lighteval/models/endpoints/inference_providers_model.py [0:0]

• 29 lines of code
• 2 McCabe index (conditional complexity)

    def __init__(self, config: InferenceProvidersModelConfig) -> None:
        """Initialize the inference client.

        Args:
            config: Configuration object containing model and provider settings
        """
        self.model_info = ModelInfo(
            model_name=config.model_name,
            model_sha="",
            model_dtype=None,
            model_size=-1,
        )
        self.model_name = config.model_name
        self.provider = config.provider
        self.generation_parameters = config.generation_parameters

        self.API_MAX_RETRY = 5
        self.API_RETRY_SLEEP = 3
        self.API_RETRY_MULTIPLIER = 2
        self.pairwise_tokenization = False
        self.semaphore = asyncio.Semaphore(config.parallel_calls_count)  # Limit concurrent API calls

        self.client = AsyncInferenceClient(
            provider=self.provider,
            timeout=config.timeout,
            proxies=config.proxies,
            bill_to=config.org_to_bill,
        )
        try:
            self._tokenizer = AutoTokenizer.from_pretrained(self.model_name)
        except HfHubHTTPError:
            logger.warning(f"Could not load model's tokenizer for the model {self.model_name}.")
            self._tokenizer = None

        self.prompt_manager = PromptManager(
            use_chat_template=True, tokenizer=self.tokenizer, system_prompt=config.system_prompt
        )