Closed ryanjulian closed 6 years ago
We should bundle the non-optimization feeds to all only be evaluated once (e.g. for the diagnostics and the baseline info). Repeatedly running the TF graph wastes time, especially on GPU.
We should bundle the non-optimization feeds to all only be evaluated once (e.g. for the diagnostics and the baseline info). Repeatedly running the TF graph wastes time, especially on GPU.