New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
How to design the flops range FLOPS_MINIMUM and FLOPS_MAXIMUM to specify the desired model Flops? #59
Comments
Hi, Thanks for your interest in our project! The maximum and minimum FLOPs should be decided by your device capacity. As the supernet converge, the PB would collect the desired and excellent subnets in the predefined flops interval. Best, Hao. |
Hi, Due to the randomness in the searching process, the architectures of the final obtained subnets vary. However, you can search for the desired structure in the interval of plus or minus 100M/200M… according to the FLOPs of target sub-model, and an architecture with similar accuracy will be obtained. Best, Hao. |
Hi, To be more specific, it's strongly recommended that you should let FLOPS_MAXIMUM decide the flops of the target model and keep FLOPS_MINIMUM constantly to 0 to obtain efficient subnets, which means you should define the interval [0M, 200M] to search for tiny models, [0, 500M] for models of middle sizes and [0M, 600M] for giant models. Best, |
Thanks for your kindly response. |
Hi, I'm closing this issue, if you have any other questions, please feel free to re-open it. Best, |
Hi,
Thanks for your excellent work. As the title show, How to design the flops range FLOPS_MINIMUM and FLOPS_MAXIMUM to specify the desired model Flops? Since the flops_minimum and flops_maxmum will influence subnets and teacher network sampling, the target model 500M and 50M may have different choices?
The text was updated successfully, but these errors were encountered: