1

From my limited understanding, it's quite annoying trying to deploy models like PyTorch. So one would need something like ONNX.

But another approach like Golang with a single binary is VERY attractive. I have used Golang before and it's reasonably fast and compiles to one binary easily. For Rust, compiling to a single binary is a nightmare.

So I wonder if anyone has seen work that translates XGBoost trees to Golang and uses arrow as the data exchange format? That way, it makes deployment super easy, wouldn't it?

xiaodai
  • 630
  • 1
  • 5
  • 13

1 Answers1

1

I code up a Golang package call gotch that can take Pytorch pretrained model and do further training/finetuning or inferring. Have a look at “example” subfolder as well as Gophernotes notebook links in README.md.

sugarme
  • 111
  • 1