Rob Suderman

There is a decent amount of "why does this model do X". I doubt `i64`s actually matter. There are also some issues where we assume an operation runs on floats...

You won't need to make any changes to `iree-compile` as it already checks for the magic bits to identify bytecode vs mlir-ir. In the case you want to see the...

Ran the model through and found a new loading failure. We reuse the `tosa.reshape` that performs a multiple collapse down / back up. We should directly lower to Expand/Collapse so...

I expect there are a decent number of these kind of dynamic shape bugs in mhlo->linalg so happy to see a new dynamic-jax has a new mechanism to generate them....

@mattjj Looking further into this, this is potentially an illegal op? It is a bit weird because you are broacasting a `tensor

Could we validate with a test that the broadcast case would pass? I am a little concerned that removing the assertion will just cause runtime failures.

I don't think you need the dialect conversion framework to handle this. I think the biggest issue is what operations are supported in the MHLO reduce region. I could easily...

Yes this should be working and committed. We can decompose quantized convolutions into regular integer convolutions with additional use of reductions and average pooling layers.

> Hello， @rsuderman ，Could you give me an opportunity to solve this issue? I may not do well initially, but I will try my best to learn and complete it!...

> I would greatly appreciate any advice and guidance you can offer. I am truly thankful and especially eager to contribute to the open-source community. So the problem overall is...