Does this support using vllm to speed up generations? #25
Description
Looking at the fastlanguagemodels, there's a field for using vllm (or fast inference), however, setting it to true seems to fail for me, so I was wondering if using vllm to increase generations is supported here.