search

facebookresearch / ijepa

Official codebase for I-JEPA, the Image-based Joint-Embedding Predictive Architecture. First outlined in the CVPR paper, "Self-supervised learning from images with a joint-embedding predictive architecture."
Other
2.83k stars 358 forks source link

Automatic1111 Integration #23

Closed Ashdevz closed 1 year ago

Ashdevz commented 1 year ago

Can anybody tell me if this can be integrated in to Automatic1111?

Thanks

MRQJsfhf commented 1 year ago

Want to know if this model can run at least 4090Ti?

MidoAssran commented 1 year ago

HI @Ashdevz, happy to help support any integration you want to try!

MidoAssran commented 1 year ago

@MRQJsfhf yes I think it should. The main changes would just be to set use_bfloat16: false in the config, and to replace torch.bfloat16 on line 316 in train.py with torch.float16.

(let me know if you try this an encounter any unexpected issues)