From 96705514bd7d5ca58e8aa9ced07b4db2615201bc Mon Sep 17 00:00:00 2001 From: Xinyuan Tong <115166877+JustinTong0323@users.noreply.github.com> Date: Sat, 13 Dec 2025 20:16:04 +0000 Subject: [PATCH] fix: dpskv32 chat history processing, default drop_thinking to true (#15064) Signed-off-by: Xinyuan Tong --- python/sglang/srt/entrypoints/openai/serving_chat.py | 4 +--- 1 file changed, 1 insertion(+), 3 deletions(-) diff --git a/python/sglang/srt/entrypoints/openai/serving_chat.py b/python/sglang/srt/entrypoints/openai/serving_chat.py index 26685cf2c..5b38bf78b 100644 --- a/python/sglang/srt/entrypoints/openai/serving_chat.py +++ b/python/sglang/srt/entrypoints/openai/serving_chat.py @@ -295,9 +295,7 @@ class OpenAIServingChat(OpenAIServingBase): messages = [msg.model_dump() for msg in messages] if request.tools: messages[0]["tools"] = [tool.model_dump() for tool in request.tools] - real_input = encode_messages( - messages, thinking_mode=thinking_mode, drop_thinking=False - ) + real_input = encode_messages(messages, thinking_mode=thinking_mode) prompt_ids = self.tokenizer_manager.tokenizer.encode(real_input) else: for message in request.messages: