[DEV] Paged Attention 支持 prefill阶段连续批次 #848

Open

Assignees

opened

on Dec 25, 2025

目标版本
main

功能描述
paged attention支持任意req数量、batch内每个req任意输入长度、任意历史长度

Metadata

Assignees

Labels

No labels

No labels

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests