Describe the bug
A clear and concise description of what the bug is.

i wanted to train the the parts of IN21k-winter class , so i made the class map of IN21k-winter and i run training.

but there is the error in timm/train.py when i use the webdataset (timm/imagent-w21-wds) with class map

This error is "KEYERROR".

when i use the class map with Image Folder(dataset is original IN21k-winter, not wds) , there is not any error.

what should do i run the training scripts with class map of timm/imagenet-w21-wds?

(plz check the pytorch-image-models/timm/data/readers/reader_wds.py)

To Reproduce
Steps to reproduce the behavior:

1. if you don't have timm/imagenet-w21-wds , plz download the dataset in huggingface.
2. make a class map file (.txt) and put the class names (e.g. n03613592, n04116512, n04027706, n04367480 ....)
3. and the run the timm/train.py with class_map

Expected behavior
A clear and concise description of what you expected to happen.

i just wanna run the training scripts with class map of timm/imagenet-w21-wds

Screenshots
If applicable, add screenshots to help explain your problem.

<class map example (.txt)>

<"error">

Desktop (please complete the following information):

• OS: Ubuntu [e.g. Windows 10, Ubuntu 18.04]
• This repository version: latest version
• PyTorch version w/ CUDA/cuDNN [e.g. from conda list, 1.7.0 py3.8_cuda11.0.221_cudnn8.0.3_0]

Additional context
Add any other context about the problem here.