Controlling the variety of mid-frequency response tokens during SFT dataset construction correlates more strongly with downstream model performance than instruction-level diversity strategies.
Title resolution pending
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
-
From Macro to Micro: Probing Dataset Diversity in Language Model Fine-Tuning
Controlling the variety of mid-frequency response tokens during SFT dataset construction correlates more strongly with downstream model performance than instruction-level diversity strategies.