The pretrained weights for 'swinv2_tiny_window16_256' in timm 0.8.x (main branch)

[leyangjin]

Hi, sorry to bother you again.

In the current main branch, which is timm 0.8.x, some models' name are not the same as their names in the url dictionary and their are also not remapped.

For example, in 'timm/models/swin_transformer_v2.py', the model name is 'swinv2_tiny_window16_256'. But in 'default_cfgs' dictionary (line 679), its name has become 'swinv2_tiny_window16_256.ms_in1k', as is shown in the following code.

'swinv2_tiny_window16_256.ms_in1k': _cfg(
        hf_hub_id='timm/',
        url='https://github.com/SwinTransformer/storage/releases/download/v2.0.0/swinv2_tiny_patch4_window16_256.pth',
    ),

The model also has not been remapped in the last few lines of the file 'timm/models/swin_transformer_v2.py'. The models that have been remapped are only a few models as follows.

register_model_deprecations(__name__, {
    'swinv2_base_window12_192_22k': 'swinv2_base_window12_192.ms_in22k',
    'swinv2_base_window12to16_192to256_22kft1k': 'swinv2_base_window12to16_192to256.ms_in22k_ft_in1k',
    'swinv2_base_window12to24_192to384_22kft1k': 'swinv2_base_window12to24_192to384.ms_in22k_ft_in1k',
    'swinv2_large_window12_192_22k': 'swinv2_large_window12_192.ms_in22k',
    'swinv2_large_window12to16_192to256_22kft1k': 'swinv2_large_window12to16_192to256.ms_in22k_ft_in1k',
    'swinv2_large_window12to24_192to384_22kft1k': 'swinv2_large_window12to24_192to384.ms_in22k_ft_in1k',
})

So does it mean that if I create the model ''swinv2_tiny_window16_256.' in timm 0.8.x using the following code: timm.creat_model('swinv2_tiny_window16_256', pretrained = True), the library will not load any pretrained weights since the model's name have not been remapped to the new name 'swinv2_tiny_window16_256.ms_in1k' yet?

Thank yo so much for your help.