dreamgaussian
dreamgaussian copied to clipboard
NaN loss on 1660 super
When training the GUI shows a nan loss and the preview is empty. It reminds me of SD's behavior when using float16 on the 16XX series.
@deividmor Hi, you may try to disable fp16 inference by changing fp16
to False
here: https://github.com/dreamgaussian/dreamgaussian/blob/main/guidance/zero123_utils.py#L22C34-L22C34