calgaryml condensed-sparsity issues

calgaryml / condensed-sparsity

[ICLR 2024] Dynamic Sparse Training with Structured Sparsity

https://openreview.net/forum?id=kOBkxFRKTA

MIT License

17 stars 3 forks source link

issues

Newest

Newest Most commented Recently updated Oldest Least commented Least recently updated

Imagenet dataloaders

#27 mklasby closed 2 years ago
0
Checkpointing

#26 mklasby closed 2 years ago
0
Benchmarks

#25 mklasby closed 2 years ago
0
docker compose template and dockerfile edits for wandb agent deployment

#24 mklasby closed 2 years ago
0
Implement build script for singularity container to run on compute canada

#23 mklasby closed 1 year ago
0
Wandb sweeps implementation

#22 mklasby closed 2 years ago
0
Condensed sparse rigl

#21 mklasby closed 2 years ago
0
Step LR per Cifar-10 rigl training regime

#20 mklasby closed 2 years ago
0
[WIP] Experimental test bed, ERK sparsity dist., cosine annealing w/ linear warm up

#19 mklasby closed 2 years ago
0
Implement linear warm up for LR scheduler

#18 mklasby closed 2 years ago
0
Confirm original rigl scheduler code working here: https://github.com/calgaryml/condensed-sparsity/blob/main/src/rigl_torch/RigL.py#L367

#17 mklasby closed 1 year ago
0
Develop motivation by fully training vanilla rigl and computing variance from const fan in across layers. See if we can identify that the original rigl scheudler yeilds an almost const fan in result by default

#16 mklasby closed 1 year ago
0
Modify Rigl torch dense allocation to match condensed sparisty (+/- small amount of overall connections to ensure const fan in can maintain the same % sparisty exactly)

#15 mklasby closed 1 year ago
0
Initalization concerns -> Scale initalization based on fan in per Yani's recent work

#14 mklasby closed 1 year ago
1
Add more docstrings for configurations, make sure variable names are descriptive and/or commented

#13 mklasby closed 1 year ago
0
Implement ResNet 18 for cifar10 with reduced kernel sizes suitable for cifar

#12 mklasby closed 2 years ago
0
Implement WideResNet22-2 for cifar10 -> Reproduce from rigl tensorflow code

#11 mklasby closed 2 years ago
0
docker dev fixes for nvidia gpg key issues and pinning docker base image

#10 mklasby closed 2 years ago
0
Implement more simple net for unit tests

#9 mklasby closed 1 year ago
0
rigl_constant_fan.py -> _get_grow_mask improvements

#8 mklasby closed 1 year ago
1
Convert masks for weight matricies into object

#7 mklasby closed 1 year ago
1
Implement ERK sparisty distribution option for RigL (inherited by rigl_constant_fan.py)

#6 mklasby closed 2 years ago
0
Rigl const fan

#5 mklasby closed 2 years ago
0
Rigl torch dockerfile integration

#4 mklasby closed 2 years ago
0
Separate development from production requirements

#3 yanii closed 2 years ago
1
Boilerplate

#2 mklasby closed 2 years ago
0
Isolate and refactor condensed sparsity layer

#1 yanii closed 2 years ago
0