Oscar Dowson

I'm going to close this until we have a motivation. In the short-term, the VectorNonlinearOracle/GrayBox can trivially handle these. So it's only an issue if we want to combine them...

Hmm. This is pretty complicated. I like the flatten/unflatten output and the oracle approach. But it seems hard for us to parse that into a full-space data structure if it...

@pulsipher do you have a small example? There are so many parameters for Conv2d, do we need them all?

The matrix input is a problem. I _really_ didn't want to have to get into this. It makes thinking about matrix shapes so much more complicated.

Oh sure. But I was hoping we could have some reshaping so that at the MathOptAI level all inputs and outputs were vectors. I need to experiment with some things.

Coincidentally, this came up in SCIP.jl last week: https://github.com/scipopt/SCIP.jl/issues/304#issuecomment-2846311778 We map `SCIP_STATUS_GAPLIMIT` to `MOI.OPTIMAL` ([code](https://github.com/scipopt/SCIP.jl/blob/05446cceea2a05e7e814398c446d1c695defe292/src/MOI_wrapper/results.jl#L14)). The SCIP log for this status is ``` SCIP Status : solving was interrupted [gap...

@amontoison here's one for you

It looks like it expects `find_library` to work for these exact strings: https://github.com/cvanaret/Uno/blob/5d37b660479e267a3d99c607ff69858b0443ce64/CMakeLists.txt#L82 so we likely need some changes to the CMake to let us inject `libblastrampoline` and to be...

We still get:

Failure is because we're still using `find_library` for BLAS etc: https://github.com/cvanaret/Uno/blob/d91ed34d5e25c3c16ebc34a7c5a67b9c35c2d33f/CMakeLists.txt#L149-L155