Closed jdiaz97 closed 8 months ago
Yes that should be possible. The docs are a bit sparse in an attempt to not repeat stuff already found in the docs for NaiveNASflux.
The tl;dr is that a loaded model is fully Functor compatible, so things like model_gpu = load("model.onnx") |> gpu
should just work.
I haven't tested this with an AMD gpu since I only have a Nvidia, but everything that works for a Flux.Chain
w.r.t parameter movement should work for models loaded with this package.
Hello! The docs are not clear about GPU support (maybe it's too implied? I'm not a fluent Flux user) I know Flux can support AMD in certain scenarios, could I run onnx models using this and AMD?