By approximating fine-tuning loss with a first-order Taylor expansion around the base model, the paper groups datasets and trains a small adapter ensemble that outperforms QLoRA on multi-dataset classification with minimal extra compute.
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
-
Efficient Ensemble for Fine-tuning Language Models on Multiple Datasets
By approximating fine-tuning loss with a first-order Taylor expansion around the base model, the paper groups datasets and trains a small adapter ensemble that outperforms QLoRA on multi-dataset classification with minimal extra compute.