Hi, is it possible to do offline Generation similar to the vllm batch Inference where the model is not served?

Like

Llm = sglang("path/to/llm")