ProDS picks instruction-tuning data by matching training-sample gradients to preference gradients from DPO, achieving slight gains over prior selection methods on MMLU, TYDIQA, BBH, and Alpaca-style tests.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
ProDS: Preference-oriented Data Selection for Instruction Tuning
ProDS picks instruction-tuning data by matching training-sample gradients to preference gradients from DPO, achieving slight gains over prior selection methods on MMLU, TYDIQA, BBH, and Alpaca-style tests.