From 969958ef1c9ad09ffffb48689e9cddc616b37ed4 Mon Sep 17 00:00:00 2001 From: ryantzr1 Date: Wed, 28 May 2025 02:44:55 +0800 Subject: [PATCH] fix(openai): improve tokenizer fallback and remove env_config param --- src/lighteval/models/endpoints/openai_model.py | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/src/lighteval/models/endpoints/openai_model.py b/src/lighteval/models/endpoints/openai_model.py index 6a6e20c29..af032fe58 100644 --- a/src/lighteval/models/endpoints/openai_model.py +++ b/src/lighteval/models/endpoints/openai_model.py @@ -87,7 +87,7 @@ def from_path(cls, path: str) -> "OpenAIModelConfig": class OpenAIClient(LightevalModel): _DEFAULT_MAX_LENGTH: int = 4096 - def __init__(self, config: OpenAIModelConfig, env_config) -> None: + def __init__(self, config: OpenAIModelConfig) -> None: self.client = OpenAI(api_key=config.api_key, base_url=config.base_url) self.config = config self.generation_parameters = config.generation_parameters @@ -107,7 +107,7 @@ def __init__(self, config: OpenAIModelConfig, env_config) -> None: try: self._tokenizer = tiktoken.encoding_for_model(self.model) except KeyError: - self._tokenizer = AutoTokenizer.from_pretrained(self.model) + self._tokenizer = tiktoken.encoding_for_model("gpt-4o") self.pairwise_tokenization = False def __call_api(self, prompt, return_logits, max_new_tokens, num_samples, logit_bias):