Bhchen
Bhchen
@gyc521zsc Hi, I met the same problem, have you solved it?
@xiaoxue17 I guess the default evaluation in tSNE is L2 distance, which does match sphereface.
@thestars-maker 首先确认mmcv版本问题；其次单卡 需要修改 lr 以及warmup（warmup x8）可以再试试
@thestars-maker warmup_iters=4000 lr先试试0.01 如果不行再改成0.01/8
@xinntao the same question
@aartykov hi did your problem be addressed？ I might meet the similar one, the progress bar doesn't change and no error gives.
> Models are trained in fp16 Hi, when training in fp16, have you encounterd the issue "Attempting to unscale FP16 gradients."?
> > > Models are trained in fp16 > > > > > > Hi, when training in fp16, have you encounterd the issue "Attempting to unscale FP16 gradients."? >...
> This is not solution, this will train it in float32 then instead of float16 I don't know, but the official code in diffusers for training controlnet using the same...
> Hai, I have trained supervised model just like steps you give (step 1-4), after training supervised baseline model on COCO dataset, I have run semi_dest.sh with corresponding file paths...