Almanac
technique

Randomized YaRN

techniqueactiveprovisionalrandomized-yarn-6c1a7cd5·1 events·first seen 2d ago

Aliases: Randomized YaRN

Co-occurring entities

More like this (12)

Recent events (1)

5arXiv · cs.CL·2d ago·source ↗

Randomized YaRN improves LLM length generalization for long-context reasoning

Researchers propose Randomized YaRN, a training method that combines YaRN-based positional extrapolation with randomized positional encodings and a length curriculum to improve LLM generalization to long contexts. Models trained on sequences under 8K tokens show consistent reasoning improvements on context lengths from 16K to 128K on BABILong and MRCR benchmarks. The key insight is that exposing models to out-of-distribution positional representations during short-context training enables better generalization at far longer inference-time lengths.