amanda99·ปีที่แล้ว·discussYou would use a KV cache to cache a significant chunk of the inference work.