Hey, I’ve been tweaking transformer pruning to cut inference time without losing accuracy—figured the trade‑off curves might interest you.