[Feature]: Support for Paged Attention

[potatoZhx]

🚀 The feature, motivation and pitch
Support Paged Attention for efficient KV Cache management in large language model inference. Inspired by virtual memory in operating systems, this technique divides KV Cache into multiple small, non-contiguous blocks, improving GPU memory utilization and throughput.

Alternatives
No response

Additional context
No response

Before submitting a new issue...

• Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the documentation page, which can answer lots of frequently asked questions.