search

keras-team / keras

Deep Learning for humans
http://keras.io/
Apache License 2.0
61.91k stars 19.45k forks source link

Regarding documentation for RegNets #15895

Closed AdityaKane2001 closed 1 year ago

AdityaKane2001 commented 2 years ago

@fchollet

#15702 was merged recently in dd73626. I wanted to add the performance metrics to the applications table in keras.io. However, I am not able to spin up a VM of the given specifications^1.

Could you please clarify how to go about this?

/cc @sayakpaul [^1]: CPU: AMD EPYC Processor (with IBPB) (92 core) - Ram: 1.7T - GPU: Tesla A100

qlzh727 commented 2 years ago

Triage notes: We should provide a user accessible way to produce performance benchmarks. Or we need to at least have it available internally to produce the number for user.

sayakpaul commented 2 years ago

I think it might be useful to put up a guide for benchmarks for contributors.

qlzh727 commented 2 years ago

Thanks @sayakpaul. Please feel free to write up a guide for how to setup this env.

On a side note, we are going to explore a setup for keras cv and nlp to benchmark the performance for applications.

sayakpaul commented 2 years ago

I think the Keras team should write to reduce the chances of errors, inconsistencies, etc.

But if you could share some references on how the team usually goes about it, I can take a dig.

For benchmarking performance (not the throughput, just accuracy), I usually do something like this:

https://github.com/sayakpaul/ViT-jax2tf/tree/main/i1k_eval.