TypeError("__init__() got an unexpected keyword argument 'aa_layer'")

[Kshitij09]

The above error is thrown for several skresnet architectures listed below:

1. skresnet18,
2. skresnet34
3. skresnet50
4. skresnet50d
5. skresnext50_32x4d

I ran a simple utility script to generate this list:

import timm
from fastprogress import progress_bar
from collections import defaultdict

failed = defaultdict(list)
for m in progress_bar(timm.list_models()):
  try: timm.create_model(m)
  except Exception as e:
    failed[e].append(m)

Any thoughts on putting this in CI or github actions?

[rwightman]

@Kshitij09 thanks, should have caught that.. normally run through the models but I was being lazy and though it wasn't needed since I'd just done it ;) .. definitely have thought about using a proper CI, at least test all model creation, feed one batch, etc

[Kshitij09]

@rwightman I've been exploring Github Actions lately and was looking for use-case to try that out. Should I try to implement one for this repository?

It'll take time but would be a good learning curve for me 🙂

[rwightman]

@Kshitij09 I have also not looked much into Github actions, but there is value in getting some sort of CI going. If you want to learn it and get a basic CI setup going on a fork, I'll bring it in if you get it to the point where it's able to at least check creating each model and maybe check a dummy forward pass works with non nans/inf.

[rwightman]

@Kshitij09 FYI @michalwols submitted a PR last night for basically the first stage, you might want to take a look re learning. I'm open to additions/enhancements in the future.

[Kshitij09]

@rwightman This looks great. I'll work on enhancements (if any). Looks like current pipeline is failing due to some issue. Will look after that.

[rwightman]

I think the EfficientNet-L2 model is taking too long (or too much mem), it was failing the user set timeout at 60s, after I increased to 5 min it looks like it's being forcefully killed by Actions. Quickest path is probably to exclude L2 since it's an absolute monster of a model.

…

On Thu, May 7, 2020 at 12:48 PM Kshitij09 ***@***.***> wrote: @rwightman <https://github.com/rwightman> This looks great. I'll work on enhancements (if any). Looks like current pipeline is failing due to some issue. Will look after that. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#139 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABLQICDOJVCRGB7JQBKOJBLRQMGABANCNFSM4MY7MHGQ> .