elixir-nx / ortex

ONNX Runtime bindings for Elixir
MIT License
134 stars 16 forks source link

Implement reshape, squeeze, and concatenate for Ortex.Backend #11

Closed mortont closed 10 months ago

mortont commented 1 year ago

Having these will reduce the number of Nx.backend_transfer calls needed when using Ortex with autoregressive models. There are likely others, similar to https://github.com/elixir-nx/ortex/issues/8 that will make this more ergonomic.

pbowyer commented 11 months ago

Looks like this issue can be closed @mortont as #12 is merged?

mortont commented 11 months ago

I still need to add concatenate before closing this, should be pretty straightforward. PRs welcome as well!

mortont commented 10 months ago

With #22 closed, I'm going to close this issue. If we want to support concatenating tensors of differing types I'd suggest a new issue. Right now, that would be a seldom (if ever) use case for Ortex since concatenation is typically only used during Nx.Serving.batched_run on tensors of the same type.