Jordan Fix

Hi @ksaurabh-cadence, it sounds like a reasonable plan. I am not sure how many others may be interested, but we'd be happy to review/land it if upstreamed.

We have some initial support for Predication, which would allow us to skip unnecessary computation in an unrolled network. That would probably be the shortest path to a highly performant...

@ajayanto We already have basic LSTM support, just not as a single node (it's currently directly implemented via its component pieces instead of as a Node that's lowered; see `Function::createLSTM()`)....

@ajayanto You don't need to provide an implementation for `fwdLSTMInst` in order to get a profile. We will want to lower the LSTMNode to its subnodes for the Interpreter backend,...

@hgarg5072 For historical reasons we didn't have an `LSTMNode` due to lowering/quantization-related issues, but those issues have been resolved, so it could be added now. And yes we should also...

Hi @ksaurabh-cadence -- I don't think much has changed here, actually. I still think the easiest path forward for now would be to allow e.g. PyTorch (or whatever is driving...

>Am I going to need 3 devices or 5 devices here? Not sure if I know your exact question -- you want to run on a single device right? You...

>I was wondering if there are 5 devices needed here or 3 or just 1. I don't want to clone F1, but call the same F1 again and again without...

Hi @yurivict -- are you using `DLLVM_DIR` during CMake to direct it to the correct llvm build? Might be worth a try to do that. Also not sure if we've...

@et-nivard Thanks for investigating this. I'd be totally fine with making SMAPE the default. My only issue is that, while some tests as you point out will now fail, others...