Socrates

I use the amp builtin pytorch and it works well. ```python from torch.cuda.amp import GradScaler, autocast if args.amp: scaler = GradScaler() for e in range(args.epoch): for idx, images in enumerate(dataloader):...

Do I need to do a test to verify this modification? If this modification is effective, I can submit a PR and test report. @lucidrains

Yep, if the target model uses train mode, the statics of BN doesn't matter since it will never be used. And in this implementation it also use train mode. But...

Indeed the implementation of magic animate is equivalent to the "concat then split" operation of animateanyone. The "query" don't need to be concated, concating the query and then get the...

> You can retrain the last layer of the animatediff from zero initialization. @guoqincode But in stage1, the color is already wrong. In stage one the color will random whitening...

> Hello @luyvlei, the results are indeed pretty decent. Could you please share what steps do I need to follow to get such results? Thanks so much in advance. Just...

> video 1 looks pretty good 👍 ！how many steps you trained the stage 1 ？ I trained 13000 steps but still can‘t control the pose :( . I have...

> > > Hello @luyvlei, the results are indeed pretty decent. Could you please share what steps do I need to follow to get such results? Thanks so much in...

> hi sir, could you tell me do you change any part of the train_hack.py? I trained 1 stage in 8 A100 with 30000 steps, but the face is terrible,...

In my experience, the jellyfin use client cookies to control the default bitrate limit. When I clean the cookies of my chrome after change the default bitrate limit to 10Mbps,...