is there any difference in efficiency/storage if i use a local file or a URL for classification inputs? e.g. does the local file get read directly or a copy is made, and/or does the URL version stay in memory or get spooled to disk somewhere?
is there any real benefit/difference to using batches when doing strictly cpu-only classifications? i know you've said with GPU work that batches are loaded into GPU memory as a whole, but I didn't know if there was any tangible benefit to batching requests in CPU-only modes.