accelerate

docs: clarify multi-process non-tensor batch limitation

#4117

Open