🚀NEW LABGetting Started with Claude AgentsStart lab
Memory

Giraffe

First page
Giraffe
Paper summary

A family of context-extended Llama and Llama 2 models, along with an empirical study of context-extension techniques.

Ask this paper

Key points
01

Extended contexts: Fine-tuned models with 4K, 16K, and 32K context windows, providing ready-to-use open long-context variants.

02

Technique comparison: Systematically compares context-extension methods including positional interpolation, truncation strategies, and attention scaling.

03

Practitioner insights: Reports practical findings on which techniques preserve downstream quality at extended contexts - useful for anyone building long-context applications.

04

Context-extension recipe: The lessons from Giraffe fed directly into the recipes that would culminate in YaRN and similar approaches later that year.

Every Monday
Get next week’s papers.
Subscribe on Substack