The open_clip_pytorch_model.bin 3.94gb file and Kaggle

[Ortange]

Hi,

I'm using Kaggle and each time I launch the notebook, it download a 3.94gb file, I think it is the clip model for SD 2.1.

The problem is that it takes 30 minutes to Kaggle to download that file, I don't know why, it downloads very fast all the time exept for that file and it's 30 minutes real minutes not exagerating.

I tried to find the place where that file is put once downloaded but found nothing in the Kaggle working folder.

I downloaded that file "open_clip_pytorch_model.bin", it weights 3.94gb, do you know where to put it in the Kaggle working folder please ?

Thanks.

[DrUmranAli]

have you tried this thread, #5108

I had the same issue and manually downloaded the files to fix. according to the suggested fix(1/2 way down in the thread)

1. a binary file "open_clip_pytorch_model.bin" renamed to "9a78ef...".

/root/.cache/huggingface/hub/models--laion--CLIP-ViT-H-14-laion2B-s32B-b79K/blobs/9a78ef8e8c73fd0df621682e7a8e8eb36c6916cb3c16b291a082ecd52ab79cc4

2. a text file contains "58a1e0...".

/root/.cache/huggingface/hub/models--laion--CLIP-ViT-H-14-laion2B-s32B-b79K/refs/main

3. a symbolic link "open_clip_pytorch_model.bin" to "../../blobs/9a78ef..."

/root/.cache/huggingface/hub/models--laion--CLIP-ViT-H-14-laion2B-s32B-b79K/snapshots/58a1e03a7acfacbe6b95ebc24ae0394eda6a14fc/open_clip_pytorch_model.bin

only issue was step 2 the file is called main and has no file extention(rename a txt file with the string)

Here, 9a78ef... is SHA256 of open_clip_pytorch_model.bin. 58a1e0... is a commit tag.

[Ortange]

Yes didn't help, these explanations are not precise enough, I don't know what to do with them. And it seems I cannot see hidden folder ".cache" in Kaggle.

All I want to know is the place where to put that "open_clip_pytorch_model.bin".

[DrUmranAli]

Sorry not familar with kaggle, but the cache folder is located

C:\Users\USERNAME\.cache\

The files are installed to C:\Users\USERNAME\.cache\huggingface\hub\models--laion--CLIP-ViT-H-14-laion2B-s32B-b79K

You need to create the huggingface\hub\models--laion--CLIP-ViT-H-14-laion2B-s32B-b79K folders if needed,

You need to have the following folders ...

blobs
refs
snapshots

in blobs, you simply rename the open_clip_pytorch_model.bin to

9a78ef8e8c73fd0df621682e7a8e8eb36c6916cb3c16b291a082ecd52ab79cc4

in refs, create a txt file called main and paste this into it

58a1e03a7acfacbe6b95ebc24ae0394eda6a14fc

then save and remame the file, deleted the .txt file extention

in snapshots, create a folder called 58a1e03a7acfacbe6b95ebc24ae0394eda6a14fc

in there create a symlink called open_clip_pytorch_model.bin to the file in the blobs folder.

It should work, I think you just need to figure out which .cache folder Kaggle is trying to use.

[Ortange]

Nope, nothing works, created the good directories, renamed the files as you said, tried different folders etc. It seems impossible to install this file manually.

I found the argument "--clip-models-path" in "shared.py" so I tried launch.py --clip-models-path="kaggle/working/CLIP/" which is where I put the 3.94gb file but still nothing, it doesn't give any error, it just downloads the file anyway.

Any other ideas ?

[Morinage]

i want to know how to download this 3.94GB file manually

[jo-xin]

https://huggingface.co/laion/CLIP-ViT-H-14-laion2B-s32B-b79K/blob/main/open_clip_pytorch_model.bin

maybe try here?