minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

fix: dpskv32 chat history processing, default drop_thinking to true (#15064)

Browse Source 
Signed-off-by: Xinyuan Tong <xinyuantong.cs@gmail.com>
This commit is contained in:
Xinyuan Tong
2025-12-13 12:16:04 -08:00
committed by GitHub
parent ab3ffd1c8e
commit 96705514bd
1 changed files with 1 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/entrypoints/openai/serving_chat.py
+1 -3
View File
@@ -295,9 +295,7 @@ class OpenAIServingChat(OpenAIServingBase):
messages = [msg.model_dump() for msg in messages] messages = [msg.model_dump() for msg in messages]
if request.tools: if request.tools:
messages[0]["tools"] = [tool.model_dump() for tool in request.tools] messages[0]["tools"] = [tool.model_dump() for tool in request.tools]
real_input = encode_messages( real_input = encode_messages(messages, thinking_mode=thinking_mode)
messages, thinking_mode=thinking_mode, drop_thinking=False
)
prompt_ids = self.tokenizer_manager.tokenizer.encode(real_input) prompt_ids = self.tokenizer_manager.tokenizer.encode(real_input)
else: else:
for message in request.messages: for message in request.messages: