Training a small predictor on hidden states from every layer of a frozen LLM improves multiple-choice QA accuracy over direct prompting and calibration baselines, approaching QLoRA fine-tuning on some benchmarks at lower cost.
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
-
InnerThoughts: Disentangling Representations and Predictions in Large Language Models
Training a small predictor on hidden states from every layer of a frozen LLM improves multiple-choice QA accuracy over direct prompting and calibration baselines, approaching QLoRA fine-tuning on some benchmarks at lower cost.