Add mode_override to the TPU embedding enqueue ops. This allows the mode to be
overridden at runtime allowing dynamic switching between inference and training modes. Not fully implemented yet. PiperOrigin-RevId: 215325071
Loading
Please sign in to comment