Saving model to file

[Ttimofeyka]

I may be wrong, but there is no writing to a file in the code and splitting the done model into checkpoints during training. Will it be added?

[Ttimofeyka]

@karpathy

[karpathy]

Yep no such thing exists. There is no reason to do this right now but we'll get there.
We're now doing multigpu bfloat16 runs, so we're doing well, probably not too far away.