bdch1234
bdch1234
Right, ok. But what I mean is that when saving/loading the cache of ConcreteFunctions, there is no way to have that done automatically? I would always need to save/restore manually?
Sure, please see below: ```python @tf.function def my_complex_operation(x): # This would then be some very complex operation, so # TF could take a while to produce the final optimized DAG....
The question I had in this thread was if something similar to the numba's cache=True exists for TensorFlow (and if not make a feature-request for it). I don't see how...
Well, then you just answered my question. Thank you. @nikitamaia, might it then be a good idea to bring back the "type:feature"-label? Or is this something you can directly say...
@nikitamaia, I have but it still not addressing the question I had. We have established that a concrete function can be saved manually if needed, but there is no functionality...
Got it. Well then thanks for providing the additional info @bhack.