New on-device engine runs AI models up to 9.9x faster without changing models

Original: Most models run on whatever inference engine they ship with. That's rarely the fastest option.

Source: x.com ↗