## P0: Model Implementation Enhance - [x] New models support: Krea-2, ERINE-Image, etc. #1088 #1074 - [ ] Separate models' codebase from diffusers or 3rd-party library, re-implement it in cache-dit as `v2`, for example: - [ ] `src/cache_dit/v2/models/transformers/transformer_flux2.py` - [ ] `src/cache_dit/v2/models/pipelines/pipeline_flux2.py` ## P1: Performance Optimization - [x] NVFP4 (non SVDQuant) Quantization Support #1090 - [ ] Flashinfer Kernel library integration - [ ] ... ## P2: API Design - [ ] Programable low level cache API (no mock required for blocks)
P0: Model Implementation Enhance
v2, for example:src/cache_dit/v2/models/transformers/transformer_flux2.pysrc/cache_dit/v2/models/pipelines/pipeline_flux2.pyP1: Performance Optimization
P2: API Design