Details
-
Sub-task
-
Status: Open
-
P2
-
Resolution: Unresolved
-
None
-
None
-
None
Description
This might be necessary for some DoFns (e.g. RunInference uses a common DoFn for different ML frameworks that expect different types).
One simple solution would be to just let DoFn implementations override the functions that infer batchconverters from typehints