Closed rick2047 closed 1 year ago
It doesn't, only GDScript.
Since the premise of this project was to train single-language models that specialize in highest-possible performance for only that language, it wouldn't make sense to include .tscn
files in the dataset.
However, it would certainly be possible to train a separate model on just .tscn
files.
I have my doubts about whether that would yield better results than just asking a GDScript model to generate code to populate a scene, and then executing that code automatically. Have no conclusive evidence either way though, past some rudimentary tests trying to populate scenes using gpt-4
.
A good thing in Godot is that the tscn format is just text. I was wondering if we can include tscn examples in the training data, we might then be able to just add things to the scene with a prompt.