[Doc]: Add documents on multimodal args

[sgsdxzy]

📚 The doc issue

Is there any documentation on what can be passed to --limit-mm-per-prompt and --mm-processor-kwargs, and their effects?
Also, is there info on what extra args are accepted in type image and video content when using the oai api?

Suggest a potential alternative/fix

No response

Before submitting a new issue...

• Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the documentation page, which can answer lots of frequently asked questions.

[DarkLight1337]

You can find the full list of engine arguments here: https://docs.vllm.ai/en/latest/serving/engine_args.html#engine-args

[sgsdxzy]

For example the doc for --mm-processor-kwargs is

Overrides for the multimodal input mapping/processing, e.g., image processor. For example: {"num_crops": 4}

but what keys and args are supported other than num_crops? Does num_crops apply to audio? And if it's model specific, where can I find which processor supports which arg?

[DarkLight1337]

It is model specific. Those arguments are forwarded to the processor obtained by AutoProcessor.from_pretrained