Recall-oriented training losses (TruncR, c-Div with alpha>1, and lambda-PR) make language models more tunable through temperature, improving precision-recall trade-offs beyond NLL plus temperature scaling.
(87) We can note that: αλ(P∥Qt θ)= ρλ+(1− ρ)λt min
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Improving Diversity in Language Models: When Temperature Fails, Change the Loss
Recall-oriented training losses (TruncR, c-Div with alpha>1, and lambda-PR) make language models more tunable through temperature, improving precision-recall trade-offs beyond NLL plus temperature scaling.