I can't release the source code - but the demo is here, will also try documenting the design.
A C++ native ML framework that follows C++ conventions and takes full advantage of its strengths.
Thinking about training a token classifier for several programming languages (e.g. C++, Go, Python) on data from LSP servers. I am curious to see:
- Is it fast enough for editors (I think it should be <1ms)
- Will it transfer understanding to languages it wasn’t trained on.