Downloading CoAtNet offline

[emanalattas]

Hello..
First of all thanks for such a great library. I need to download this model "coatnet_rmlp_2_rw_224.sw_in12k_ft_in1k" and use it offline.
How can I do this? Where can I find the model exactly?
How can I use it after downloading it?
Thanks in advance.

[emanalattas]

I found these 2 ways but I don't know how to download the model weights.
1)
model_name = 'ModelName'
model = timm.create_model(model_name, pretrained=False)
saved_weights_path = 'MyPath'
model.load_state_dict(torch.load(saved_weights_path))

model = timm.create_model(model_name,pretrained=True,pretrained_cfg_overlay=dict(file=saved_weights_path))

So, how can I save the model weights locally?

[hankyul2]

Hi @emanalattas

You can download model weight via hf_hub_download(weight_name, 'pytorch_model.bin'). And you can load the saved pytorch_model.bin file using the torch.load function. The model weight is usually saved at ~/.cache/huggingface/hub folder. Here is a code snippet.

import timm
from huggingface_hub import hf_hub_download
import torch

saved_path = hf_hub_download('timm/coatnet_rmlp_2_rw_224.sw_in12k_ft_in1k', 'pytorch_model.bin', library_name='timm', library_version=timm.__version__)
state_dict = torch.load(saved_path, map_location='cpu')

+) timm.create_model(model_name, pretrained=True) will automatically download the model weight to ~/.cache/huggingface/hub folder. So you can load the weight in offline also.

I hope this could help you. Thank you.

Hankyul

[emanalattas]

Thanks Hankyul. I appreciate your reply.
I use Windows 11 and run the code on Google Colab.
When I print the path I got:
print(saved_path)
/root/.cache/huggingface/hub/models--timm--coatnet_rmlp_2_rw_224.sw_in12k_ft_in1k/snapshots/e90722529d2c8c6778e99cdb1c3c42e45a2e6cd1/pytorch_model.bin

But I didn't find this location in my laptop.
Can you help me to figure out its location?
Thanks in advance.

[hankyul2]

Hi @emanalattas

As the Google Colab seems to prohibit browsing specific folders such as root (maybe I'm wrong), you can alternatively save the model weight to your /home/ folder and access it through the file browser.

You can download model weight to a specific folder e.g., /home/ by setting the local_dir parameter in hf_hub_download(..., local_dir='/home/') like the below code snippet.

import timm
from huggingface_hub import hf_hub_download
import torch

saved_path = hf_hub_download('timm/coatnet_rmlp_2_rw_224.sw_in12k_ft_in1k', 'pytorch_model.bin', local_dir='/home/', library_name='timm', library_version=timm.__version__)
state_dict = torch.load(saved_path, map_location='cpu')

I hope this answer could help you. Thank you.

Hankyul.

[emanalattas]

Hello @hankyul2
Yes it works. I just changed local_dir='/home/' to local_dir='/content' then I download it manually from Google Colab to my device.

Thanks a lot for your help. I appreciate it.