minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 13 Packages Projects Releases Wiki Activity

Handle unlimited tokenizer context lengths (#37054)

Browse Source 
Co-authored-by: Baizhou Zhang <sobereddiezhang@gmail.com>
This commit is contained in:
Neeraj Kashyap
2026-08-29 17:29:11 -07:00
committed by GitHub
co-authored by Baizhou Zhang
parent f60bc73c58
commit 032fe91bf1
1 changed files with 5 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/entrypoints/openai/serving_tokenize.py
+5
View File
@@ -45,6 +45,11 @@ class OpenAIServingTokenize(OpenAIServingBase):
try: try:
tokenizer = self.tokenizer_manager.tokenizer tokenizer = self.tokenizer_manager.tokenizer
max_model_len = getattr(tokenizer, "model_max_length", -1) 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: if request.messages is not None:
token_ids = self._tokenize_chat_request(request) token_ids = self._tokenize_chat_request(request)