Skip to content

[RFC] v0.5.0 Unified Cache Backend #216

@DefTruth

Description

@DefTruth

v0.5.0 Features Trace

  • Design a Unified Cache Backend Abstract: Cache Indicator (When to cache) + Cache calibrator (How to cache)
  • New algorithms support: FoCa, CG-Taylor (PaddlePaddle), Group Block Cache & token-wise TaylorSeer (from cache-dit), HiCache, DiCache, etc.
  • More new models support
  • Official readthedocs.io
  • Tests CI
  • compatible with cpu offload
  • ...

Metadata

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

Issue actions