Crusoe models

Run model inference with fast time-to-first-token, low latency, limitless throughput, and resilient scaling.