Solutions for Streaming with Access to Token Probabilities in llama.cpp?

[SinlinLi]

Hello,

I'm currently building a chatbot using llama.cpp, and I want the model to respond multiple times to a single user query. My plan is to decide whether the model should continue its response based on the probability of the token. For low-latency responses, I'm using stream=True, but I noticed that when streaming is enabled, I cannot get the logprobs to display.

Is there any way to enable logprobs while streaming responses? If not, are there any potential workarounds to achieve both streaming and access to token probabilities, particularly for ?

Thank you for your assistance!