[BUG] vLLMWrapper prompt_logprobs kwarg should not be a boolean

[albertbou92]

Describe the bug

I am might be missing sth here because the SOTA code works fine, but in theory the vllm.SamplingParams prompt_logprobs parameter can not be a boolean as it is defined here in the class vLLMWrapper:
https://github.com/pytorch/rl/blob/main/torchrl/modules/llm/policies/vllm_wrapper.py#L213

Instead, the prompt_logprobs should be None by default and 1 if generate is not provided.

Seen in the vllm docs: https://docs.vllm.ai/en/v0.7.0/api/inference_params.html

Checklist

• I have checked that there is no similar issue in the repo (required)
• I have read the documentation (required)
• I have provided a minimal working example to reproduce the bug (required)

[albertbou92]

alright it was just the env variable VLLM_USE_V1=0

[vmoens]

Yeah it's a bit annoying and I'm not sure how to move forward with that!