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
From #443, encoders/decoders are used to handle serialization of fields with datatypes like np.ndarray and torch.Tensor. This is not longer well supported in pydantic V2. I am guessing we could modernize by using @field_serializer for much of this or wrapping the array objects.
Activity
nikitakuklev commentedon Jul 16, 2026
I have a custom implementation of serialization + compression in apsopt (ala b64zstd:(encoded array data)), will try port over as custom serializers.
roussel-ryan commentedon Jul 20, 2026
This would be a great contribution @nikitakuklev thanks for taking this on