Why Supervised Fine-Tuning Fails to Learn: A Systematic Study of Incomplete Learning in Large Language Models
For practitioners fine-tuning LLMs, this work reveals that standard SFT can leave systematic gaps in learning, necessitating fine-grained diagnosis of what and why models fail to learn.
The paper identifies and systematically studies the Incomplete Learning Phenomenon (ILP) in supervised fine-tuning of LLMs, where models fail to correctly reproduce a subset of their training data even after convergence. Experiments across Qwen, LLaMA, and OLMo2 show ILP is widespread and heterogeneous, with aggregate metrics masking persistent unlearned subsets.
Supervised Fine-Tuning (SFT) is the standard approach for adapting large language models (LLMs) to downstream tasks. However, we observe a persistent failure mode: even after convergence, models often fail to correctly reproduce a subset of their own supervised training data. We refer to this behavior as the Incomplete Learning Phenomenon(ILP). This paper presents the first systematic study of ILP in LLM fine-tuning. We formalize ILP as post-training failure to internalize supervised instances and demonstrate its prevalence across multiple model families, domains, and datasets. Through controlled analyses, we identify five recurrent sources of incomplete learning: (1) missing prerequisite knowledge in the pre-trained model, (2) conflicts between SFT supervision and pre-training knowledge, (3) internal inconsistencies within SFT data, (4) left-side forgetting during sequential fine-tuning, and (5) insufficient optimization for rare or complex patterns. We introduce a diagnostic-first framework that maps unlearned samples to these causes using observable training and inference signals, and study several targeted mitigation strategies as causal interventions. Experiments on Qwen, LLaMA, and OLMo2 show that incomplete learning is widespread and heterogeneous, and that improvements in aggregate metrics can mask persistent unlearned subsets. The findings highlight the need for fine-grained diagnosis of what supervised fine-tuning fails to learn, and why.