snimu / rebasin Goto Github PK
View Code? Open in Web Editor NEWApply methods described in "Git Re-basin"-paper [1] to arbitrary models --- [1] Ainsworth et al. (https://arxiv.org/abs/2209.04836)
License: MIT License
Apply methods described in "Git Re-basin"-paper [1] to arbitrary models --- [1] Ainsworth et al. (https://arxiv.org/abs/2209.04836)
License: MIT License
Hi,
I'm having a weird situation, where when paired with the diffusers SDXL model, the program makes it ~130 layers deep of initialize_parallel_paths with CPU memory usage climbing to 80G after a few minutes (both models passed into the program were on GPU),
weirdly taking close to no CPU or GPU utilization while running.
Thanks for the amazing project, I have an inquiry regarding the visualization feature mentioned in the tutorial.
What does this mean? Can you elaborate a little further please?
Thanks!
ParallelPaths(
LinearPath( LinearPath(
DefaultModule( DefaultModule(
module.type: Conv2d module.type: Conv2d
input.shape: [(128, 64, 56, 56)] input.shape: [(128, 64, 56, 56)]
output.shape: [(128, 64, 56, 56)] output.shape: [(128, 256, 56, 56)]
weight.in_dim.permutation: None weight.in_dim.permutation: None
weight.out_dim.permutation.shape: 64 weight.out_dim.permutation.shape: 256
) )
OneDimModule( OneDimModule(
module.type: BatchNorm2d module.type: BatchNorm2d
input.shape: [(128, 64, 56, 56)] input.shape: [(128, 256, 56, 56)]
output.shape: [(128, 64, 56, 56)] output.shape: [(128, 256, 56, 56)]
weight.in_dim.permutation.shape: 64 weight.in_dim.permutation.shape: 256
weight.out_dim.permutation.shape: 64 weight.out_dim.permutation.shape: 256
) )
DefaultModule( )
module.type: Conv2d |
input.shape: [(128, 64, 56, 56)] |
output.shape: [(128, 64, 56, 56)] |
weight.in_dim.permutation.shape: 64 |
weight.out_dim.permutation.shape: 64 |
) |
|
OneDimModule( |
module.type: BatchNorm2d |
input.shape: [(128, 64, 56, 56)] |
output.shape: [(128, 64, 56, 56)] |
weight.in_dim.permutation.shape: 64 |
weight.out_dim.permutation.shape: 64 |
) |
|
DefaultModule( |
module.type: Conv2d |
input.shape: [(128, 64, 56, 56)] |
output.shape: [(128, 256, 56, 56)] |
weight.in_dim.permutation.shape: 64 |
weight.out_dim.permutation.shape: 256 |
) |
|
OneDimModule( |
module.type: BatchNorm2d |
input.shape: [(128, 256, 56, 56)] |
output.shape: [(128, 256, 56, 56)] |
weight.in_dim.permutation.shape: 256 |
weight.out_dim.permutation.shape: 256 |
) |
|
(please refer to attached txt file for full result):
A declarative, efficient, and flexible JavaScript library for building user interfaces.
๐ Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. ๐๐๐
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google โค๏ธ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.