How to extend inference ctx by yarn out of training data?

[FNsi]

Let's say, the case is,
I want to throw ggml-quants.c to model, but the length is beyond 131072.

How to properly using YARN Handle this situation?