transmuteai / chipnet Goto Github PK

Hi, amazing work guys. I was wondering if you have some code readily available to extract the fine-tuned pruned sub-model out of the checkpoint model, so that it can be used by simple inference scripts?

Hi,

Is it possible to release the code for your network slimming implementation which is reported in the paper. The original code seems to have problems with large pruning ratio.

Thanks.

Hi,

We just came across your work, it was amazing!

We have a question about the ResNet pruning that when we run the ChipNet code and get these results.

Test Acc: 0.7316
Total Trainable Params: 317198
Channels pruned: 80.00%
Parameters pruned: 81.63%

Test Acc: 0.7556
Total Trainable Params: 853033
Channels pruned: 60.00%
Parameters pruned: 50.59%

We saw that the output channels for the blocks are not the same, so how can we get a smaller model with it?
Did you do padding to make sure they are the same?

First of all, thanks for your work.
But it seems that the ChipNet just uses masks, there is no real removal parameter.
How to get the actual model compression ratio?

Please refer to the CIFAR-10, Resnet-50 architecture in the table 2 of paper.

When volume budget is 12.5 %, number of parameter is 2.8% and FLOPs is 5.1% in Table 2.

However, the my reproduced result after pruning step is the number of parameter is 12.9% and FLOPs is 16.9 %
while the volume budget is 12.5%.

I follow the hyperparameter setting provided in the appendix.

Would you share train-set validation-set split information and pretrained networks to reproduce the results in the paper??

There is an ambiguous naming convention for the finetuned model with transfer masks technique.