vm_loss config
zheedong opened this issue · comments
Hi, can you let me know what is correct training config in stage 1 vm_loss?
Is it in StableDiffusionHead,
- drop_prob = 0.1
- noise_offset = 0.0
- input_perturbation = 0.1
- snr_gamma = 5.0
right?
Where can I see proper training config?
Thank you.
Hi @zheedong,
No. The recommendation is from the Hugginface team instead of us. We use the default configurations as follows:
- drop_prob = None
- noise_offset = 0.0
- input_perturbation = 0.0
- snr_gamma = None
The configuration can be modified, but we have not. So, the configuration is just the default.
Hi @zheedong,
Is it clear now or you need some more help?
Ah sorry for late.
No thank yo for your help.