Adam Gibson

Just Google jmh harness. We have some samples in the contriv/nd4j-benchmarks directory too. On Wed, Jan 1, 2025, 16:18 Massimiliano Zattera ***@***.***> wrote: > That's why I was asking you...

@mzattera that's fine. The main thing we need to do is confirm my theory about the data loading being the bottleneck and what happens if you let it warm up...

@diavole hmm the basics seem ok. Are you building on your own local system then deploying to another one? Note: why do you need the UI after the model is...

@sdyura looking at this it's actually probably best to fully delete this model saver. You'd definitely always want proper snapshots of training. A score should also be actively evaluated with...

That's fine. As I said it's still not a priority right now. You clearly know where the bug is. Since it's not a low level c++ style bug and seems...

Odd..we have so many test cases for LSTMs. Do you know if this is using keras 3?

@mzattera what knime version are you using? I'm sure they'll tell you what version they're using from that. I can at least try to find it on my end. Given...

@mzattera https://docs.knime.com/2018-12/deep_learning_installation_guide/index.html#known-issues it looks like it's keras 2.x. We should be able to load that then...let me double check on my end.

@mzattera it might already be fixed in the current master. I need to test against M2.1 and SNAPSHOT. Unfortunately that's only available in the repo atm due to a major...

Yeah that's actually all I need! If you can just check with m2.1 that would be great. If not I will check later. On Mon, Dec 30, 2024, 17:09 Massimiliano...