Gitlab & Hugging Face load support - #319
Conversation
…agger into wk9874/models/hf_load_support
|
@praksharma also one for you to look at to see how best to integrate it with your HF model recipes stuff! I'm hoping that effectively the dummy YOLO model implementation I have in |
…agger into wk9874/models/hf_load_support
…agger into wk9874/models/hf_load_support
…agger into wk9874/models/hf_load_support
…agger into wk9874/models/hf_load_support
samueljackson92
left a comment
There was a problem hiding this comment.
Looks good to me and works!
samueljackson92
left a comment
There was a problem hiding this comment.
Actually I just spotted that i get the weights file dumped in the local folder where I ran toktagger? Surely it should be in a cache folder?
This was an error in the temp YOLO implementation rather than the actual loading code - Have fixed it, can you double check it works now pls? |
It looks to similar to #326 (comment). Not sure if you tested it on NVIDIA GPU. |
|
Working nicely now. Just merged the dev to get the |
Initial commit adding Gitlab & HF Load Support
Not 100% happy with the UI elements, lots of repetition and I'm sure it can be tidied.
Gitlab loading doesn't work on UKAEA instance since its on too old of a version, I hope this will be bumped fairly soon so I can finish that part & test
Hugging Face loading works, added a temporary YOLOv8 implementation into
api/models/temp.pywhich can be used to test. This whole file should be removed before this PR is mergedCan take an initial look at the implementation if you get time @samueljackson92 / @abdullah-ukaea , and then I can respond to comments when I'm back from AL
TODOs:
Safe tensors limitingMove Gitlab / HF settings to their own subheadingsTests