If a model uses the same tensor for both an input and an output then this can result in data loss and memory corruption. This should not happen. PiperOrigin-RevId: 332522916 Change-Id: If0905b142415a9dfceaf2d181872f2a8fb88f48a
432 B
432 B