Comment by swiftcoder

This has to be a distilled model? I don't see any route to 200 tokens/s without serious datacenter hardware