A Mechanistic Investigation of Supervised Fine Tuning

• URL: http://arxiv.org/abs/2605.11426v1
• Date: Tue, 12 May 2026 02:22:48 GMT
• Title: A Mechanistic Investigation of Supervised Fine Tuning
• Abstract summary: Supervised Fine-Tuning (SFT) leaves the model's activation geometry largely undisturbed.<n> projecting both sets of activations through a Sparse Autoencoder pretrained on the base model reveals that the underlying sparse latents diverge significantly.
• Score: 0.0
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: The cosine similarity between a large language model's hidden activations before and after Supervised Fine-Tuning (SFT) remains very high. This, at first glance, suggests that SFT leaves the model's activation geometry largely undisturbed. However, projecting both sets of activations through a Sparse Autoencoder (SAE) pretrained on the base model reveals that the underlying sparse latents diverge significantly. We introduce a novel investigative pipeline which utilizes these pretrained SAEs as a high-resolution diagnostic tool to mechanistically investigate the drivers of this representational divergence. Through our analytical pipeline, we discover task-specific and layer-specific distributions of the precise semantic features that are systematically altered during supervised fine-tuning. We additionally identify a layer-wise update profile specific to safety alignment. All code, experimental scripts, and analysis files associated with this work are publicly available at: https://github.com/ruhzi/sae-investigation.

Related papers

• Beyond the Black Box: Interpretability of Agentic AI Tool Use [0.0]
  We introduce a mechanistic-interpretability toolkit built on Sparse Autoencoders and linear probes.<n>The framework reads model states before each action and infers both whether a tool is needed and how consequential the next tool action is likely to be.<n>We train the probes on multi-step trajectories from the NVIDIA Nemotron function-calling dataset and apply the same workflow to GPT-OSS 20B and Gemma 3 27B models.
  arXiv  Detail & Related papers  (2026-05-07T19:47:30Z)
• Dynamic analysis enhances issue resolution [53.50448142467294]
  DAIRA (Dynamic Analysis-enhanced Issue Resolution Agent) is an automated repair framework that embeds dynamic analysis into the agent's reasoning cycle.<n> Driven by a Test Tracing-Driven methodology, DAIRA utilizes lightweight monitors to extract critical runtime data.<n>When powered by Gemini 3 Flash Preview, DAIRA establishes a new state-of-the-art (SOTA) performance, achieving a 79.4% resolution rate on the SWE-bench Verified dataset.
  arXiv  Detail & Related papers  (2026-03-23T14:48:54Z)
• Step-Level Sparse Autoencoder for Reasoning Process Interpretation [48.99201531966593]
  Large Language Models (LLMs) have achieved strong complex reasoning capabilities through Chain-of-Thought (CoT) reasoning.<n>We propose step-level sparse autoencoder (SSAE), which serves as an analytical tool to disentangle different aspects of LLMs' reasoning steps into sparse features.<n> Experiments on multiple base models and reasoning tasks show the effectiveness of the extracted features.
  arXiv  Detail & Related papers  (2026-03-03T14:25:02Z)
• Toward Faithful Explanations in Acoustic Anomaly Detection [21.487734134424187]
  Interpretability is essential for user trust in real-world anomaly detection applications.<n>We compare a standard autoencoder (AE) with a mask autoencoder (MAE) in terms of detection performance and interpretability.<n>MAE consistently provides more faithful and temporally precise explanations, suggesting a better alignment with true anomalies.
  arXiv  Detail & Related papers  (2026-01-19T02:16:37Z)
• Every Step Counts: Decoding Trajectories as Authorship Fingerprints of dLLMs [63.82840470917859]
  We show that the decoding mechanism of dLLMs can be used as a powerful tool for model attribution.<n>We propose a novel information extraction scheme called the Directed Decoding Map (DDM), which captures structural relationships between decoding steps and better reveals model-specific behaviors.
  arXiv  Detail & Related papers  (2025-10-02T06:25:10Z)
• Interpretable Anomaly Detection in Encrypted Traffic Using SHAP with Machine Learning Models [0.0]
  This study aims to develop an interpretable machine learning-based framework for anomaly detection in encrypted network traffic.<n>Models are trained and evaluated on three benchmark encrypted traffic datasets.<n> SHAP visualizations successfully revealed the most influential traffic features contributing to anomaly predictions.
  arXiv  Detail & Related papers  (2025-05-22T05:50:39Z)
• Isolation Forest in Novelty Detection Scenario [0.0]
  novelty detection focuses on identifying previously unseen patterns after training solely on regular data.<n>In this work, we explore the Half-Space Tree (HST) algorithm, originally proposed for streaming anomaly detection.<n>We propose a novel theoretical modification to adapt it specifically for novelty detection tasks.
  arXiv  Detail & Related papers  (2025-05-13T12:21:53Z)
• Open-Set Deepfake Detection: A Parameter-Efficient Adaptation Method with Forgery Style Mixture [81.93945602120453]
  We introduce an approach that is both general and parameter-efficient for face forgery detection.<n>We design a forgery-style mixture formulation that augments the diversity of forgery source domains.<n>We show that the designed model achieves state-of-the-art generalizability with significantly reduced trainable parameters.
  arXiv  Detail & Related papers  (2024-08-23T01:53:36Z)
• Interpreting Attention Layer Outputs with Sparse Autoencoders [3.201633659481912]
  Decomposing model activations into interpretable components is a key open problem in mechanistic interpretability. In this work we train SAEs on attention layer outputs and show that also here SAEs find a sparse, interpretable decomposition. We show that Sparse Autoencoders are a useful tool that enable researchers to explain model behavior in greater detail than prior work.
  arXiv  Detail & Related papers  (2024-06-25T17:43:13Z)
• DeTra: A Unified Model for Object Detection and Trajectory Forecasting [68.85128937305697]
  Our approach formulates the union of the two tasks as a trajectory refinement problem. To tackle this unified task, we design a refinement transformer that infers the presence, pose, and multi-modal future behaviors of objects. In our experiments, we observe that ourmodel outperforms the state-of-the-art on Argoverse 2 Sensor and Open dataset.
  arXiv  Detail & Related papers  (2024-06-06T18:12:04Z)
• ComPtr: Towards Diverse Bi-source Dense Prediction Tasks via A Simple yet General Complementary Transformer [71.82644727907146]
  We propose a novel $underlineComP$lementary $underlinetr$ansformer, $textbfComPtr$, for diverse bi-source dense prediction tasks.<n>ComPtr treats different inputs equally and builds an efficient dense interaction model in the form of sequence-to-sequence on top of the transformer.
  arXiv  Detail & Related papers  (2023-07-23T15:17:45Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.