how to count MACs

[ZTzxj]

I'd like to ask you exactly how you calculated the macs for stable diffusion with calflops, I'm getting errors when I use it, sorry to bother you!

[LukeLIN-web]

Same question. SD-XL. MrYxJ/calculate-flops.pytorch#23
MrYxJ/calculate-flops.pytorch#12 (comment)
author said The calflops don't support diffuse model for the time being.

[llCurious]

Same question.
Any ideas? @HaozheLiu-ST @WentianZhang-ML . Thx a lot!

[HaozheLiu-ST]

Hi, sorry for the late reply... I'm swamped in these months...

Thanks for your interest in our paper. We modified the diffusers and calflops libraries to align their FLOPs calculations. This adjustment was done directly within the original code in a rather ad hoc manner, so we did not organize or archive this portion after completing the project. The FLOPs calculation primarily relies on the implementation in pytorch_ops.py.

[ZTzxj]

Ok ok, thank you very much for your reply, we've solved the problem, but thanks for the detailed answer! 发件人：Haozhe ***@***.***> 发送日期：2025-02-26 18:28:33 收件人：HaozheLiu-ST/T-GATE ***@***.***> 抄送人：ZTzxj ***@***.***>,Author ***@***.***> 主题：Re: [HaozheLiu-ST/T-GATE] how to count MACs (Issue #21) Hi, sorry for the late reply... I'm swamped in these months... Thanks for your interest in our paper. We modified the diffusers and calflops libraries to align their FLOPs calculations. This adjustment was done directly within the original code in a rather ad hoc manner, so we did not organize or archive this portion after completing the project. The FLOPs calculation primarily relies on the implementation in pytorch_ops.py. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: ***@***.***> HaozheLiu-ST left a comment (HaozheLiu-ST/T-GATE#21) Hi, sorry for the late reply... I'm swamped in these months... Thanks for your interest in our paper. We modified the diffusers and calflops libraries to align their FLOPs calculations. This adjustment was done directly within the original code in a rather ad hoc manner, so we did not organize or archive this portion after completing the project. The FLOPs calculation primarily relies on the implementation in pytorch_ops.py. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: ***@***.***>

[LukeLIN-web]

Ok ok, thank you very much for your reply, we've solved the problem, but thanks for the detailed answer!

发件人：Haozhe @.>
发送日期：2025-02-26 18:28:33
收件人：HaozheLiu-ST/T-GATE @.>
抄送人：ZTzxj @.>,Author @.>
主题：Re: [HaozheLiu-ST/T-GATE] how to count MACs (Issue #21)
Hi, sorry for the late reply... I'm swamped in these months...
Thanks for your interest in our paper. We modified the diffusers and calflops libraries to align their FLOPs calculations. This adjustment was done directly within the original code in a rather ad hoc manner, so we did not organize or archive this portion after completing the project. The FLOPs calculation primarily relies on the implementation in pytorch_ops.py.
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you authored the thread.Message ID: @.>
HaozheLiu-ST left a comment (HaozheLiu-ST/T-GATE#21)
Hi, sorry for the late reply... I'm swamped in these months...
Thanks for your interest in our paper. We modified the diffusers and calflops libraries to align their FLOPs calculations. This adjustment was done directly within the original code in a rather ad hoc manner, so we did not organize or archive this portion after completing the project. The FLOPs calculation primarily relies on the implementation in pytorch_ops.py.
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you authored the thread.Message ID: @.>

How did you solve the problem?