Support for Logprobs Output in Triton Inference Server with vLLM Backend

[RafaelXokito]

I recently deployed the Triton Inference Server with the vLLM backend. However, we have a use case where we need to obtain the logprobs (logarithmic probabilities) for the generated tokens along with the regular output.

Currently, it appears that Triton with the vLLM backend does not directly support returning logprobs as part of the response.

Is there any existing workaround or upcoming feature that would allow us to retrieve logprobs from the vLLM backend through Triton?