extending

Definition

Pushing the range of an existing capability outward. In LLM papers it most often refers to context-length extension: making a model handle inputs longer than those it was trained on. The word implies reuse rather than retraining from scratch — you take an existing checkpoint and adapt its positional scheme, usually with a modest amount of additional training, so the cost is small relative to the gain.

Explainers using this term

• How Long-Context LLMs Work — From RoPE Interpolation to Ring AttentionExtending Context Window of Large Language Models via Positional Interpolation