Logit reweighting, especially a threshold rule that boosts likely topic tokens, increases topic-vocabulary use in summaries from Gemma-2B and Llama-3-8B with little measured quality loss.
A survey on in-context learning, 2023
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
REJECT 1representative citing papers
citing papers explorer
-
Logit Reweighting for Topic-Focused Summarization
Logit reweighting, especially a threshold rule that boosts likely topic tokens, increases topic-vocabulary use in summaries from Gemma-2B and Llama-3-8B with little measured quality loss.