Manages the KV Cache with OS-style paging (blocks + block tables); enables Continuous Batching in vLLM.

What it does?

1. 96% utilization
2. No fragmentation
3. Non-contiguous memory

How it does?

1. OS memory management
2. KV Blocks
3. Block tables

References

1. https://www.youtube.com/watch?v=5ZlavKF_98U