#Smart Search CLIP model download is too slow

1 messages · Page 1 of 1 (latest)

small peak
#

I wanted to change the CLIP model for Smart Search to something multilangual. I tried it with two different models and they both run into a timeout:
(nllb-clip-large-siglip__mrl __and XLM-Roberta-Large-ViT-H-14__frozen_laion5b_s13b_b90k)
So I tried to extract the link for the files it's trying to download and it's a model.onnx file which is 1.6GB and the speed is below 100kb/s so it will take many hours which seems to be the reason it runs into a timeout
I don't know if theres a way to fix this besides hoping that their servers get faster somehow...
But what would also help me would be if someone knew where the files for the models are placed and if I can put them there manually after I downloaded everything with jdownloader or something like that.
I'll add the error message in the comments
edit: I'm using an immich_machine_learning docker on my windows machine that has a cuda capable GPU

red ospreyBOT
#

:wave: Hey @small peak,

Thanks for reaching out to us. Please follow the recommended actions below; this will help us be more effective in our support effort and leave more time for building Immich immich.

References

Checklist

  1. :blue_square: I have verified I'm on the latest release(note that mobile app releases may take some time).
  2. :blue_square: I have read applicable release notes.
  3. :blue_square: I have reviewed the FAQs for known issues.
  4. :blue_square: I have reviewed Github for known issues.
  5. :blue_square: I have tried accessing Immich via local ip (without a custom reverse proxy).
  6. :blue_square: I have uploaded the relevant logs, docker compose, and .env files, making sure to use code formatting.
  7. :blue_square: I have tried an incognito window, disabled extensions, cleared mobile app cache, logged out and back in, different browsers, etc. as applicable

(an item can be marked as "complete" by reacting with the appropriate number)

If this ticket can be closed you can use the /close command, and re-open it later if needed.

small peak
tropic iris
#

What country are you located in?

small peak
#

germany

tropic iris
#

hm, haven't heard any issues

small peak
#

maybe I can try a few other models until I find one that doesn't download slowly

#

or try using a vpn

#

this is the link I used to download the model.onnx file:

#

https://cdn-lfs-us-1.huggingface.co/repos/3c/58/3c5813c94f07ab26517e1e4962c4576e0ceeb23e85e8df24c1d246c43d2ac8ad/d4f893a034e973e42d6dfed3884503a034c6db544f637b0cb58bcb139f080ec0?response-content-disposition=inline%3B+filename*%3DUTF-8''model.onnx%3B+filename%3D"model.onnx"%3B&Expires=1724260235&Policy=eyJTdGF0ZW1lbnQiOlt7IkNvbmRpdGlvbiI6eyJEYXRlTGVzc1RoYW4iOnsiQVdTOkVwb2NoVGltZSI6MTcyNDI2MDIzNX19LCJSZXNvdXJjZSI6Imh0dHBzOi8vY2RuLWxmcy11cy0xLmh1Z2dpbmdmYWNlLmNvL3JlcG9zLzNjLzU4LzNjNTgxM2M5NGYwN2FiMjY1MTdlMWU0OTYyYzQ1NzZlMGNlZWIyM2U4NWU4ZGYyNGMxZDI0NmM0M2QyYWM4YWQvZDRmODkzYTAzNGU5NzNlNDJkNmRmZWQzODg0NTAzYTAzNGM2ZGI1NDRmNjM3YjBjYjU4YmNiMTM5ZjA4MGVjMD9yZXNwb25zZS1jb250ZW50LWRpc3Bvc2l0aW9uPSoifV19&Signature=arC4oNHnItP6mxCXcMHCdvUEeveEw3INvT1VVH7F85dNKjcWWkEQn9~dkryWGRNlO6OWRRg~wEVyxhyHjtWaqbWwS8m8DXwsIItcRiMQZ1sMxHI6~0uSguQ8mqQIPnPaQVt0mk9l~NoxlQYmN3fiJxMv43KLmOjD6NdnycHD2A6iDlBG1mtO5s2Assl~-7M7jC5RpQY5krrKvOPSGlD~xFdpzZXM8AsEIBAnG5Py9QSOcZzXTOqoIwCVK7wUFGT54UbUqbxkUTrBJLK2mJv~rGP4M968PRmhvhUEoT6Q5uyU-BFXNSc8kdkBKSbQ1ed44G4H7IXMHvY~OB~DqGRxAg__&Key-Pair-Id=K24J24Z295AEI9

#

(I extracted it from the error message)

small peak
#

It seems like it was just yesterday this slow

#

but I still had half a model in the cache and it wouldn't clean up the unfinished download on it's own

#

for anyone who runs into this problem in the future, I deleted the model in the /cache/clip/ Folder and then it restarted the download