Instruction-tuned Llama 3.1 controls word count far better than the base model, and attribution scores point to later layers, but the scoring rule mishandles outputs that are too short.
In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), ACL 2023, Toronto, Canada, July 9-14, 2023, pages 1317–1328
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2025 1verdicts
REJECT 1representative citing papers
citing papers explorer
-
How Instruction-Tuning Imparts Length Control: A Cross-Lingual Mechanistic Analysis
Instruction-tuned Llama 3.1 controls word count far better than the base model, and attribution scores point to later layers, but the scoring rule mishandles outputs that are too short.