Lab to support speculative decoding on the spark
Thread with lots of performance numbers for inference in single and cluster mode: