Adapter Merging Reactivates Latent Reasoning Traces: A Mechanism Analysis
On the map: LLM reasoning → Test-time compute and efficient thinking
Abstract (arXiv)
Large language models fine-tuned via a two-stage pipeline (domain adaptation followed by instruction alignment) can exhibit non-trivial interference after adapter merging, including the re-emergence of explicit reasoning traces under strict decoding. We study this phenomenon in medical LLM settings using lightweight, reproducible measurements of trace leakage and instruction-following behavior. Beyond marker-based proxies, we introduce a marker-forbidden, answer-only evaluation and define a correctness-based direction that does not rely on surface markers; a rank-1 logit-space intervention along this direction modulates decision distributions and improves multiple-choice accuracy beyond random-direction controls at sufficiently large intervention strength. We further provide layer-wise geometric evidence that domain and instruction adapters induce partially misaligned update directions, and present a proof-of-concept geometry-aware merge that can reduce leakage and/or improve accuracy in a toy setting. Our results characterize boundary conditions of trace leakage and provide practical diagnostics and interventions for safer adapter merging.
Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in Test-time compute and efficient thinking, one of 17 directions in LLM reasoning. Builds on 6 of its 9 references that are in the map. First posted on arXiv 26 Jan 2026.
Builds on
6 of 9 references.
References · 6
- 2022Training language models to follow instructions with human feedbackcited by 15,971
- 2019Parameter-Efficient Transfer Learning for NLPcited by 4,236
- 2021Finetuned Language Models Are Zero-Shot Learnerscited by 3,780
- 2023QLoRA: Efficient Finetuning of Quantized LLMscited by 3,418
- 2022Self-Instruct: Aligning Language Models with Self-Generated Instructionscited by 2,724
- 2020Don't Stop Pretraining: Adapt Language Models to Domains and Taskscited by 1,565
Built on by
Not cited yet.
Similar, not cited
Close in wording, no citation either way
- 2025In Their Own Words: Reasoning Traces Tailored for Small Models Make Them Better Reasoners
- 2026Reasoning-Trace Collapse: Evaluating the Loss of Explicit Reasoning During Fine-Tuning
- 2026Transcoder Adapters for Reasoning-Model Diffing
- 2026The Shape of Overthinking: Backtracking Bursts in Long Reasoning Traces
- 2026Compliance vs. Sensibility: On the Reasoning Controllability in Large Language Models
- 2026Diagnosis Is Not Prescription: Linguistic Co-Adaptation Explains Patching Hazards in LLM Pipelines
- 2026When Reasoning Traces Become Performative: Step-Level Evidence that Chain-of-Thought Is an Imperfect Oversight Channel
- 2025The Thinking Spectrum: An Empirical Study of Tunable Reasoning in LLMs through Model Merging