ericproffitt / TopicModelsVB.jl

A Julia package for variational Bayesian topic modeling.
Other
81 stars 8 forks source link

Cannot include in package using current Julia because of OpenCL dependency #40

Open neelsmith opened 3 years ago

neelsmith commented 3 years ago

I'm trying to include TopicModelsVB in a module built with Julia 1.6. I can use it and pass unit tests, but if I try to register the package with juliahub, it fails to compile because of the dependency on OpenCL.

Likewise, even though the functions I need from TopicModelsVB work fine in unit tests, if I try Pkg.test("OpenCL") locally, that fails.

ericproffitt commented 3 years ago

Hi Neel,

I'm not sure I can offer much support if the problem is with the OpenCL package. You may need to request that JuliaHub manually register your package, or otherwise coordinate with them on troubleshooting this issue.

The way I register this package is by opening an issue like you see here: https://github.com/ericproffitt/TopicModelsVB.jl/issues/20.

Auto-registration fails because of OpenCL, but then someone is able to manually merge my package.

neelsmith commented 3 years ago

Thanks! I've just started using TopicModelsVB recently, but already love it. Great work!

For now, I've isolated the sections of my module using TopicModelsVB in a separate module which I'm just adding from the repo's URL, but your suggestion would be a nice alternative to that. It looks like OpenCL isn't compatible with Julia 1.x and it looks like there hasn't been a lot of activity on it recently. Do you have any ideas about moving away from it?

Thanks for the quick reply, and for the great code.

Neel

On Tue, Jul 13, 2021 at 12:39 PM Eric Proffitt @.***> wrote:

Hi Neel,

I'm not sure I can offer much support if the problem is with the OpenCL package. You may need to request that JuliaHub manually register your package, or otherwise coordinate with them on troubleshooting this issue.

The way I register this package is by opening an issue like you see here:

20 https://github.com/ericproffitt/TopicModelsVB.jl/issues/20.

Auto-registration fails because of OpenCL, but then someone is able to manually register my package.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/ericproffitt/TopicModelsVB.jl/issues/40#issuecomment-879237892, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAQ2FBQODW3DYHGCOMD7RMDTXRT4XANCNFSM5AJJSQZA .

ericproffitt commented 3 years ago

I've just started using TopicModelsVB recently, but already love it. Great work!

Thanks!

So as far as I know OpenCL is compatible with Julia 1.x. I think their run tests may be out of date, however the GPU topic models work fine on my MacBook Pro using Julia version 1.6.1.

I don't have any plans to move away from OpenCL.jl, as it's the only OpenCL package available for Julia, and OpenCL is the only cross-platform framework available for GPU programming. At some point I would like to write equivalent algorithms for CUDA, however I don't currently have access to an Nvidia GPU.

neelsmith commented 3 years ago

Thanks -- maybe I'm looking at out of date docs for OpenCL. I wish I did have hardware for CUDA or Apple Metal...

On Wed, Jul 14, 2021 at 11:04 AM Eric Proffitt @.***> wrote:

Thanks! I've just started using TopicModelsVB recently, but already love it. Great work!

Thanks!

So as far as I know OpenCL is compatible with Julia 1.x. I think their run tests may be out of date, however the GPU topic models work fine on my MacBook Pro.

I don't have any plans to move away from OpenCL.jl, as it's the only OpenCL package available for Julia, and OpenCL is the only cross-platform framework for GPU programming. At some point I would like to write equivalent GPU algorithms for CUDA and Apple Metal, however I don't currently have access to either of these GPU platforms.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/ericproffitt/TopicModelsVB.jl/issues/40#issuecomment-879970827, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAQ2FBSH7WX7PBJODA5TWCLTXWRRRANCNFSM5AJJSQZA .