[Usage]: prompt_logprompt from endpoint

[basma-b]

Your current environment

I want to get the logprobs from a vLLM endpoint on the prompt + answer in order to evaluate the LLM on selective task. How can I do that?

curl --location URL/v1/chat/completions \
--header "Content-Type: application/json" \
--data '{
	"model": "model_name",
	"echo": true,
	"messages": [
		{
			"role": "user", "content": "hello"}
		],
	"logprobs": true,
	"top_logprobs": 1
}'

I am using this code but I get only the logprobs of the answer. Can anyone help please?

How would you like to use vllm

I want to run inference of a [specific model](put link here). I don't know how to integrate it with vllm.