mlsanityData to 14 Oct 2026

A Layer-wise Analysis of Supervised Fine-Tuning

arXiv 2604.11838 ↗First posted 12 Apr 2026cs.LG cs.AIcited by 2

Not on the map. Why?

Abstract (arXiv)

While critical for alignment, Supervised Fine-Tuning (SFT) incurs the risk of catastrophic forgetting, yet the layer-wise emergence of instruction-following capabilities remains elusive. We investigate this mechanism via a comprehensive analysis utilizing information-theoretic, geometric, and optimization metrics across model scales (1B-32B). Our experiments reveal a distinct depth-dependent pattern: middle layers (20%-80%) are stable, whereas final layers exhibit high sensitivity. Leveraging this insight, we propose Mid-Block Efficient Tuning, which selectively updates these critical intermediate layers. Empirically, our method outperforms standard LoRA up to 10.2% on GSM8K (OLMo2-7B) with reduced parameter overhead, demonstrating that effective alignment is architecturally localized rather than distributed. The code is publicly available at https://anonymous.4open.science/r/basesft.

Key facts. Cited by 2 papers in the map (data through 14 Oct 2026). Builds on 22 of its 37 references that are in the map. First posted on arXiv 12 Apr 2026.

Builds on

22 of 37 references.

References · 22

  1. 2022Training language models to follow instructions with human feedbackcited by 15,971
  2. 2021Training Verifiers to Solve Math Word Problemscited by 10,109
  3. 2023Judging LLM-as-a-Judge with MT-Bench and Chatbot Arenacited by 8,626
  4. 2020Measuring Massive Multitask Language Understandingcited by 7,963
  5. 2021Finetuned Language Models Are Zero-Shot Learnerscited by 3,780
  6. 2016Pointer Sentinel Mixture Modelscited by 3,096
  7. 2022Locating and Editing Factual Associations in GPTcited by 2,691
  8. 2023LIMA: Less Is More for Alignmentcited by 1,100
  9. 2023Steering Llama 2 via Contrastive Activation Additioncited by 1,085
  10. 2019BERT Rediscovers the Classical NLP Pipelinecited by 1,059
  11. 2023The Unlocking Spell on Base LLMs: Rethinking Alignment via In-Context Learningcited by 257
  12. 2024MathScale: Scaling Instruction Tuning for Mathematical Reasoningcited by 144
10 more
  1. 2024Collective Constitutional AI: Aligning a Language Model with Public Inputcited by 123
  2. 2024Keeping LLMs Aligned After Fine-tuning: The Crucial Role of Prompt Templatescited by 96
  3. 2024A Closer Look at the Limitations of Instruction Tuningcited by 77
  4. 2023From Language Modeling to Instruction Following: Understanding the Behavior Shift in LLMs after Instruction Tuningcited by 56
  5. 2023Instructed to Bias: Instruction-Tuned Language Models Exhibit Emergent Cognitive Biascited by 53
  6. 2023Do Models Really Learn to Follow Instructions? An Empirical Study of Instruction Tuningcited by 49
  7. 2024Learning or Self-aligning? Rethinking Instruction Fine-tuningcited by 43
  8. 2023Instruction-tuning Aligns LLMs to the Human Braincited by 40
  9. 2025SAIF: A Sparse Autoencoder Framework for Interpreting and Steering Instruction Following of Language Modelscited by 34
  10. 2023LIMIT: Less Is More for Instruction Tuning Across Evaluation Paradigmscited by 14

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 2.

2 papers.

  1. 2026Rotation-Preserving Supervised Fine-Tuningcited by 3
  2. 2026Crafting Reversible SFT Behaviors in Large Language Modelscited by 0