search

Q-Future / Q-Align

③[ICML2024] [IQA, IAA, VQA] All-in-one Foundation Model for visual scoring. Can efficiently fine-tune to downstream datasets.
https://q-align.github.io
Other
293 stars 19 forks source link

How to train without huggingface_hub and offline? #13

Closed RollingWang closed 8 months ago

RollingWang commented 8 months ago

Thanks for your interesting job, and I want to reproduce it in my server offline, but it's confused about how to avoid using huggingface_hub?

teowu commented 8 months ago

If you are in Mainland China, please try to use hf-mirror.com; You can also pre-download the pre-trained weights and save locally, and change here to your local dir: LOAD='MAGAer13/mplug-owl2-llama2-7b'