0.0.44

ema model should only be updated on main

0.0.43

saving checkpoint

0.0.42

saving and loading trainer

0.0.41

setup basic recon validation

0.0.40

better initial scale for xcit linear attention

0.0.39

latest accelerate

0.0.38

fully wire up accelerate

0.0.37

preliminary training loop

0.0.36

just copy over a optimizer factory fn

0.0.35

apply transform to each frame of video