Hacker News Reader

They do:

> For specific use cases, we support other languages, including Java, Erlang, Haskell, and Go. These languages are currently not widely supported outside of specific use cases.

> Probably the data size is so high that the only way to survive there is to write everything in low level C++ or Rust.

Except they say: “For data science, ML applications, and Instagram, Python continues to be the language of choice, and we continue to invest in the experience with this ecosystem.”

For research you use Python or whatever. For production you compile your models to whatever their proprietary equivalent to TorchScript and ONNX is, which is surely executed by hyperoptimized C++/Rust/Cuda code.

PyTorch came out of Meta AI and I'm pretty sure they are still the primary contributors, so I assume they use TorchScript.

Same here. I was hoping to see some Scala… perhaps for data processing/streaming/engineering.

The build tool they use, Buck, is written in Java, although a rewrite of it in Rust is in the works.

Buck2 in rust is already functional