Properly support batched/non-batched with vllm/llama.cpp

Properly support batched/non-batched with vllm/llama.cpp #148