Supervised fine-tuning of LLMs often fails to fully internalize all training instances due to five recurring causes including missing prerequisites and data conflicts, as diagnosed via a new framework across multiple models.
Liang Li, Qisheng Liao, Meiting Lai, Di Liang, and Shangsong Liang
3 Pith papers cite this work. Polarity classification is still indexing.
3
Pith papers citing it
citation-role summary
other 1
citation-polarity summary
fields
cs.CL 3years
2026 3verdicts
UNVERDICTED 3roles
other 1polarities
unclear 1representative citing papers
E-GRM triggers CoT reasoning in generative reward models only when parallel generations show high uncertainty, reducing inference cost and raising accuracy on reasoning benchmarks via a hybrid regression-ranking scorer.
The paper claims a selective fine-tuning method that identifies and freezes core parameters to mitigate catastrophic forgetting in LLMs while improving domain adaptation, shown in experiments with GPT-J and LLaMA-3.
citing papers explorer
-
Why Supervised Fine-Tuning Fails to Learn: A Systematic Study of Incomplete Learning in Large Language Models
Supervised fine-tuning of LLMs often fails to fully internalize all training instances due to five recurring causes including missing prerequisites and data conflicts, as diagnosed via a new framework across multiple models.
-
Reason Only When Needed: Efficient Generative Reward Modeling via Model-Internal Uncertainty
E-GRM triggers CoT reasoning in generative reward models only when parallel generations show high uncertainty, reducing inference cost and raising accuracy on reasoning benchmarks via a hybrid regression-ranking scorer.
-
Efficient Task Adaptation in Large Language Models via Selective Parameter Optimization
The paper claims a selective fine-tuning method that identifies and freezes core parameters to mitigate catastrophic forgetting in LLMs while improving domain adaptation, shown in experiments with GPT-J and LLaMA-3.