Gender bias in Pythia-6.9b grows sharply after about 80k training steps even as general performance improves, and stopping earlier could trade 1.7% LAMBADA accuracy for a large fairness gain.
On the impact of machine learning randomness on group fairness
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.CL 1years
2025 1verdicts
CONDITIONAL 1roles
background 1polarities
unclear 1representative citing papers
citing papers explorer
-
Fairness Dynamics During Training
Gender bias in Pythia-6.9b grows sharply after about 80k training steps even as general performance improves, and stopping earlier could trade 1.7% LAMBADA accuracy for a large fairness gain.