← Back

YaRN

technique 1 mention from 1 sources

Yet another RoPE extensioN - a technique for extending the context length capabilities of models using rotary position embeddings.

1

sources

Mentioned by

All mentions

Sebastian Raschka mentioned ✓ High confidence
"They had a YaRN extension and there was some custom scaling there, and I couldn't quite match these things."

Attribution: Sebastian mentions YaRN as part of the technical implementation challenges in OLMo 3