Cerebras has built a platform for push-button training of large language models that can accelerate time to insights without having to orchestrate across a large cluster of small devices.