Skip to content

Latest commit

 

History

History

DeepCache

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
extension
extension
 
 
sd
sd
 
 
sdxl
sdxl
 
 
svd
svd
 
 
README.md
README.md
 
 
__init__.py
__init__.py
 
 
flops.py
flops.py
 
 

README.md

Code

Explanation of Code for SD and SDXL

We make some modifications to the code for SD and SDXL. Here we use the code of SD as an example to highlight the key changes:

  1. pipeline_stable_diffusion.py:

Added sample_from_quad_center function (Lines 86-96) for non-uniform sampling, allowing selection of a specific number of steps from the total inference steps. Modified the __call__ of StableDiffusionPipeline (Lines 726-762) to incorporate feature caching in the outermost loop.

  1. pipeline_utils.py: Incorporated code (Lines 317-318) to load the model class specifically designed for DeepCache.

  2. unet_2d_block.py: Adjusted the forward functions of CrossAttnDownBlock2D and CrossAttnUpBlock2D to either reuse or cache features.

  3. unet_2d_condition.py: Altered the forward function (Lines 957-1142) to also reuse or cache features.