We finish off our short series on SLM optimization with the third entry, focused on batching by length instead of looping ...