Awesome work, but I am genuinely curios about the use cases where the 200ms init time being a problem?
Lambda is far more economic given the memory requirements, but we recently moved to Rust+Candle to shave off ~300ms on cold starts as the lag could be really jarring.