0.1.1

Add serving logic. You can launch bloom-based LLM speculative sampling as a server.

0.1.0

Demonstrate Speculative Sampling using bloom 560m and 7b1 models.
Support KV Cache Optimization.
Only works for batch size as 1.