serge icon indicating copy to clipboard operation
serge copied to clipboard

bug: Allow loading .gguf and .bin files

Open mhussaincov94 opened this issue 1 year ago • 3 comments

would it be possible to add support to run gguf moddles like this? https://huggingface.co/TheBloke/Kunoichi-7B-GGUF I have been testing, and simply changeing the file extention from .gguf to .bin loads it ok. but having support for the .gguf extention would help. image I am using: ghcr.io/serge-chat/serge:main debian unstable with all updates installed. docker docker compose installed and up-to-date. using chromium from the debian repo. thank you for considdering my request.

Checklist
  • [X] Modify docker-compose.yml ✓ https://github.com/serge-chat/serge/commit/f408cac78b60b4326a35cfdbc7edb45c8d2aa6e7 Edit
  • [X] Running GitHub Actions for docker-compose.ymlEdit
  • [X] Modify charts/serge/values.yaml ✓ https://github.com/serge-chat/serge/commit/105f0fdbe5e86c701da83f930596b84acf425937 Edit
  • [X] Running GitHub Actions for charts/serge/values.yamlEdit
  • [X] Create scripts/model_loader.py ✓ https://github.com/serge-chat/serge/commit/eadd93ca5a5426f65cefa2db839a8dd7bf0da108 Edit
  • [X] Running GitHub Actions for scripts/model_loader.pyEdit
  • [X] Modify README.md ✓ https://github.com/serge-chat/serge/commit/761e0dcfce2f0c60f7b8dbf9653837974803939a Edit
  • [X] Running GitHub Actions for README.mdEdit

mhussaincov94 avatar Feb 05 '24 12:02 mhussaincov94

That works already the .bin is just how we rename the files. Serge supports GGUF.

gaby avatar Feb 05 '24 13:02 gaby

hi, they don't work. if added with the file extention .gguf I have just tested by adding the moddle to the default moddles directory, restarting the container and it is not showing up.

mhussaincov94 avatar Feb 05 '24 13:02 mhussaincov94

You have to add them with .bin, else they wont get loaded into the UI.

gaby avatar Feb 06 '24 00:02 gaby