[BUG] can't download the pretrained efficientnet pruned models

huggingface / pytorch-image-models

The largest collection of PyTorch image encoders / backbones. Including train, eval, inference, export scripts, and pretrained weights -- ResNet, ResNeXT, EfficientNet, NFNet, Vision Transformer (ViT), MobileNetV4, MobileNet-V3 & V2, RegNet, DPN, CSPNet, Swin Transformer, MaxViT, CoAtNet, ConvNeXt, and more

https://huggingface.co/docs/timm

Apache License 2.0

31.56k stars 4.71k forks source link

[BUG] can't download the pretrained efficientnet pruned models #1465

Closed mxer closed 2 years ago

mxer commented 2 years ago

Describe the bug It seems that efficientnet pruned models download links url='https://imvl-automl-sh.oss-cn-shanghai.aliyuncs.com/darts/hyperml/hyperml/job_45403/outputs/effnetb1_pruned_9ebb3fe6.pth', are invalid

To Reproduce Steps to reproduce the behavior: 1.use timm.create_model('efficientnet_b1_pruned', pretrained=True), got "urllib.error.HTTPError: HTTP Error 403: Forbidden"

Screenshots A7403F8E-446D-47c5-80BF-384A8E9B5087

rwightman commented 2 years ago

@mxer hmm, indeed, looks like some other models are impacted too. I'll move all those to timm github release assets from my cache in the next day or so...

rwightman commented 2 years ago

If you're in a hurry, I went through all of the files from aliyuncs.com,

did a quick pickle security scan (somethign I was going to do for outside hosted weights)
confirmed match of sha256 hash (if there was one)
cleaned up some naming for consistency
re-extracted the state dict
re-calculated sha256 and save checkpoint with current pytorch format
uploaded to https://github.com/rwightman/pytorch-image-models/releases/tag/v0.1-tresnet (you can download / update manually using these url if you are in a rush)

I have yet to update the model default cfg url in the code, but will get to that tomorrow or wed and cut a new release...

mxer commented 2 years ago

Cool, Thx!