You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I ran the training scripts (using DiT-S/8 by default) successfully, and the loss curve is shown below.
However, the sampled results (also using the default sampling parameters with DiT-XL/2 modified to DiT-S/8) are random noises.
Is that because the model is weak?
Could you please provide a recommended setting (hyper-params like the model arch, compression, etc) that we should start with?
The text was updated successfully, but these errors were encountered:
I ran the training scripts (using DiT-S/8 by default) successfully, and the loss curve is shown below.
However, the sampled results (also using the default sampling parameters with DiT-XL/2 modified to DiT-S/8) are random noises.
Is that because the model is weak?
Could you please provide a recommended setting (hyper-params like the model arch, compression, etc) that we should start with?
The text was updated successfully, but these errors were encountered: