You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
This repository was archived by the owner on Jul 7, 2023. It is now read-only.
It seems that you saved the checkpoint with a pre-v1.3.0 version of Tensor2tensor and now you try to load it with v1.3.0 or later version. This is not possible, see the 1.3.0 release notes.
In general, I don't recommend mixing t2t versions - even if the checkpoint format is compatible, there may be other subtle differences, e.g. different optimal hyperparameters (e.g. as reported in #529).
BTW: If you don't use pastebin for long log output, please use Markdown (three backtics) to mark the code, so it renders with an extra scrollbar.
It is useful also to always report your tensor2tensor (and tensorflow) version.
No, I tried with the same versions. I just ran, stopped and ran again without any delays between these actions.
Anyway, now with current versions installed by pip it works great. So it is not actual anymore.
t2t_trainer fails to restore saved checkpoints. I run it first time and broke at some point, then run the same command again.
Here is the full output:
The text was updated successfully, but these errors were encountered: