Does sglang support offline benchmark tools?

[NiuMa-1234]

I'm using sglang's offline inference for RL, and I found that there seems to be only online benchmark tools. Is there any offline tool (and how to use it) or would this feature be supported in the future?

My offline call is like this:

    loop = asyncio.get_event_loop()
    output = loop.run_until_complete(
    self._engine.async_generate(
    prompt=None,  # because we have already convert it to prompt token id
    sampling_params=self.sampling_params,
    return_logprob=True,
    input_ids=idx_list,
    image_data=image_list,
                    )
                )

An online benchmark example:

python -m sglang.bench_serving \
    --backend sglang \
    --model Qwen/Qwen3-VL-235B-A22B-Instruct \
    --dataset-name random-image \
    --output-details \
    --profile