Open nwittler opened 2 years ago
Shapes and types of tensorflow variables needs to be consistent and correct across the whole codebase.
The current workaround are casts and reshapes to fix inconsistencies. This is performance poison and needs to be corrected.
Die the slow death of garbage performance.
Describe the missing feature
Shapes and types of tensorflow variables needs to be consistent and correct across the whole codebase.
Describe the solution you'd like
The current workaround are casts and reshapes to fix inconsistencies. This is performance poison and needs to be corrected.
Describe alternatives you've considered
Die the slow death of garbage performance.