What is the meaning of rwand in21ft1k in the model name?

[klae01]

import timm
timm.list_models("*efficientnetv2*")

['efficientnetv2_l',
 'efficientnetv2_m',
 'efficientnetv2_rw_m',
 'efficientnetv2_rw_s',
 'efficientnetv2_rw_t',
 'efficientnetv2_s',
 'efficientnetv2_xl',
 'gc_efficientnetv2_rw_t',
 'tf_efficientnetv2_b0',
 'tf_efficientnetv2_b1',
 'tf_efficientnetv2_b2',
 'tf_efficientnetv2_b3',
 'tf_efficientnetv2_l',
 'tf_efficientnetv2_l_in21ft1k',
 'tf_efficientnetv2_l_in21k',
 'tf_efficientnetv2_m',
 'tf_efficientnetv2_m_in21ft1k',
 'tf_efficientnetv2_m_in21k',
 'tf_efficientnetv2_s',
 'tf_efficientnetv2_s_in21ft1k',
 'tf_efficientnetv2_s_in21k',
 'tf_efficientnetv2_xl_in21ft1k',
 'tf_efficientnetv2_xl_in21k']

I found a model that included rw while searching the model using timm, and I thought that the training method or the data used would be different, so I came into this repository while searching to find out the difference.
Is rw the initials of your name? Or, does it have a different meaning like reweighted?

Also, I would appreciate it if you could tell me what 'in21ft1k' means.

[rwightman]

@klae01

rw (my initials) means it is not the official model architecture but my 'guess' before the official code was released, they work well though and some people have reported the rw_s especially is easy to work with for fine-tuning

in21ft1k ==> pretrained on ImageNet21k , fine-tuned on 1k.