Kai Xu

@MikeInnes I need some help from you here. I cannot run tests on my local and saw the error message below. What did I do wrong? ```julia WARNING: could not...

@MikeInnes This works on my local now.

Looking forward to https://github.com/JuliaGPU/CUDAnative.jl/pull/334 > ## try > # Build failed > * [ci/gitlab/trying](https://gitlab.com/JuliaGPU/CuArrays.jl/pipelines/70047176) It seems that some tests not related to this PR fails.

@maleadt Any idea what I should do for this PR?

Thanks for bumping this up @torfjelde. It would be great if you can port this PR to CUDA.jl - let me know if you need any help.

@yebai We avoided this problem by constructing a new vector of dual from the real parts, which only happens when we need to use the vector in AD: ```julia vi[range[i]]...

How is the current situation of ReverseDiff.jl working with GPUArrays.jl?

> Is that roughly correct? Yes. And also yes for NUTS regarding "is_accept = whether the sample is accepted (is always true?)"? > BTW a bit confusing that there's both...

AHMC for sure outputs `log_density` ([here](https://github.com/TuringLang/AdvancedHMC.jl/blob/master/src/trajectory.jl#L227) for static and [here](https://github.com/TuringLang/AdvancedHMC.jl/blob/master/src/trajectory.jl#L767) for NUTS), and Turing should have warpped them into chains as they are declared [here](https://github.com/TuringLang/Turing.jl/blob/master/src/inference/Inference.jl#L124-L143). So it looks like a...

I see. Is the only edge case being sparse matrices? If so maybe we can implement something similar to `result_type` and deal with them explicitly?