Yuge Zhang
Yuge Zhang
Thanks for pointing this out. You are right that latency-loss is no longer supported in current version of ProxylessNAS. We are planning a systematic support for latency-aware NAS, coming in...
Interesting. But unfortunately `trainng_step()` is a pytorch-lightning API, and therefore there's not much we can do...
`nnictl webui nas` has been deprecated long ago. If you intend to use it, please downgrade your nni.
With v2.8, you have to do it on your own. With the upcoming next release (v2.9), there are some metrics in tensorboard that you can monitor.
Temporarily fixed to 0.1.2 via #1625.
Hi @haoren3696, hyperopt 0.2 has updated its API and NNI hasn't updated its implementation accordingly. We believe that this is a low priority issue. Please tell us why this is...
@haoren3696, thanks for letting us know that you are experiencing such issue. We will put it on the agenda, hopefully in the next release. Meanwhile, pip has a `--no-dependencies` option....
# v2.9 Release ## Neural Architecture Search ### Test items - [x] Run DARTS and NAS-Bench-201 (multiple algo) example in https://github.com/microsoft/nni/pull/5052 @QuanluZhang - [x] View and resume of NAS experiment....
# Release note ## Neural Architecture Search * New tutorial of model space hub and one-shot strategy ([tutorial](https://nni.readthedocs.io/en/latest/tutorials/darts.html)) * Add pretrained checkpoints to AutoFormer ([doc](https://nni.readthedocs.io/en/latest/reference/nas/search_space.htm.retiarii.hub.pytorch.AutoformerSpace)) * Support exporting state dict...