Anyone Successfully Finetune 2b Model on Own Datasets?

[2ezisez2]

I found it really hard to finetune d32 + 2b on my own dataset T.T
Mean bit accuracy is around 72%, and the results still get artifacts.
If anyone succeed, would you please share your experience & setting detail with me? Really appreciate!!

[Paul-Liyb]

Same question......

Seems so hard to get to 80%

[asdcaszc]

Hi,

Do you know how to only finetune the VAE rather than VAE and AR model? I cannot find any codes from the readme.

[Paul-Liyb]

Hi,

Do you know how to only finetune the VAE rather than VAE and AR model? I cannot find any codes from the readme.

You can refer to their BitVAE repo, https://github.com/FoundationVision/BitVAE.git