How to disable <think> tag on stream mode request?
#4
by
celsowm
- opened
Hi !
I tried to include "enable_thinking" on request body calling a llama-server using "enable_thinking" but no effect:
const requestBody = {
messages: [
{
role: 'user',
content: 'who are you?',
}
],
temperature: 1.0,
top_p: 1.0,
model: 'unsloth/Qwen3-4B-GGUF',
enable_thinking: false,
stream: true,
};
Even when using softblock "/no_think" still returning an empty tag
Any hint?
This is as expected. Dealing with CoT and empty thinking is the client's responsibility.
celsowm
changed discussion status to
closed