There's currently no standard because there's no one objective best way of handling prompt syntax.

There's some libraries which use the OpenAI API syntax as a higher-level abstraction, but for the lower-level precompiled binaries used in in this post that's too much.

llamafile can provide an abstraction, but ultimately it boils down to how the model was trained and/or fine-tuned.