From 032fe91bf1710bcffa9baedfa6dec96178b1ec1b Mon Sep 17 00:00:00 2001 From: Neeraj Kashyap Date: Sat, 29 Aug 2026 17:29:11 -0700 Subject: [PATCH] Handle unlimited tokenizer context lengths (#37054) Co-authored-by: Baizhou Zhang --- python/sglang/srt/entrypoints/openai/serving_tokenize.py | 5 +++++ 1 file changed, 5 insertions(+) diff --git a/python/sglang/srt/entrypoints/openai/serving_tokenize.py b/python/sglang/srt/entrypoints/openai/serving_tokenize.py index 6eaddc58a..4fa19a78c 100644 --- a/python/sglang/srt/entrypoints/openai/serving_tokenize.py +++ b/python/sglang/srt/entrypoints/openai/serving_tokenize.py @@ -45,6 +45,11 @@ class OpenAIServingTokenize(OpenAIServingBase): try: tokenizer = self.tokenizer_manager.tokenizer max_model_len = getattr(tokenizer, "model_max_length", -1) + # ORJSON accepts signed negative and unsigned positive 64-bit integers. + if isinstance(max_model_len, int) and not ( + -(2**63) <= max_model_len < 2**64 + ): + max_model_len = self.tokenizer_manager.model_config.context_len if request.messages is not None: token_ids = self._tokenize_chat_request(request)