On a local AI server, a prompt too big for the model gets silently truncated, and raising the window for every parallel slot is what froze our lab server....
Categories