Ultra-Fast Inference On Own Infrastructure With 1ms latency for BERT-like models | Hacker News Reader