Differentiable Faithfulness Alignment for Cross-Model Circuit Transfer

• URL: http://arxiv.org/abs/2604.24302v1
• Date: Mon, 27 Apr 2026 10:49:10 GMT
• Title: Differentiable Faithfulness Alignment for Cross-Model Circuit Transfer
• Abstract summary: We introduce DFA, a framework that transfers circuit information from a smaller source model to a larger target model.<n>We evaluate DFA on Llama-3 and Qwen-2.5 across six tasks spanning factual retrieval, multiple-choice reasoning, and arithmetic.<n>DFA consistently outperforms simple baselines and, in some settings, recovers target-model circuits with faithfulness comparable to or stronger than direct attribution.
• Score: 73.0825876244911
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Mechanistic interpretability has made it possible to localize circuits underlying specific behaviors in language models, but existing methods are expensive, model-specific, and difficult to scale to larger architectures. We introduce \textbf{Differentiable Faithfulness Alignment (DFA)}, a framework that transfers circuit information from a smaller source model to a larger target model through a learned differentiable alignment. DFA projects source-model node importance scores into the target model and trains this mapping with a soft faithfulness objective, avoiding full circuit discovery on the target model. We evaluate DFA on Llama-3 and Qwen-2.5 across six tasks spanning factual retrieval, multiple-choice reasoning, and arithmetic. The strongest results occur on Llama-3 $1$B$\rightarrow3$B, where aligned circuits are often competitive with direct node attribution and zero-shot transfer remains effective. Recovery weakens for larger source--target gaps and is substantially lower on Qwen-2.5, suggesting that transfer becomes harder as architectural and scaling differences increase. Overall, DFA consistently outperforms simple baselines and, in some settings, recovers target-model circuits with faithfulness comparable to or stronger than direct attribution. These results suggest that smaller models can provide useful mechanistic priors for larger ones, while highlighting both the promise and the limits of node-level cross-model circuit alignment.\footnote{Code is available at https://github.com/jasonshaoshun/dfa-circuits.

Related papers

• Architecture-Dependent Causal Transfer of Activation States Across Large Language Models [51.56484100374058]
  Internal language activation states can be transferred causally between different large language models.<n>Injecting projected activations into a target model during generation produces a statistically significant, preregistered causal effect.<n>We conclude that end-to-end activation-state transfer between LLMs, as currently implemented, is architecture-dependent rather than universal.
  arXiv  Detail & Related papers  (2026-08-17T09:53:27Z)
• Beyond Scale and Generation: Understanding Language Model-based Entity Matching [18.904187328683513]
  We show that model variant is critical for bi-encoders.<n>Generative matchers do not universally outperform cross-encoders.<n> larger models rely more heavily on shortcut learning.
  arXiv  Detail & Related papers  (2026-07-27T17:29:18Z)
• AnE: Pushing the Reasoning Frontier of Multimodal LLMs via Anchor Evolution [61.593935260052795]
  Post-training viaSupervised Fine-Tuning (SFT) and Reinforcement Learning (RL) is crucial for enhancing reasoning in Multimodal Large Language Models (MLLMs)<n>Existing paradigms often reach a performance bottleneck due to the limitations of static data.<n>We propose Anchor Evolution (AnE), a new paradigm that integrates truth-anchored data curation and model evolution.
  arXiv  Detail & Related papers  (2026-05-25T08:26:34Z)
• Architecture, Not Scale: Circuit Localization in Large Language Models [0.0]
  We show that grouped query attention produces circuits far more concentrated and mechanistically stable than standard multi-head attention at comparable scales.<n>The same concentration pattern holds across indirect object identification, induction heads, and factual recall.<n>These findings suggest that some architectural choices make large models more tractable to study.
  arXiv  Detail & Related papers  (2026-05-09T10:05:22Z)
• Cross-Architecture Model Diffing with Crosscoders: Unsupervised Discovery of Differences Between LLMs [4.8608047541422925]
  Crosscoders are one solution capable of cross-architecture model diffing but have only ever been applied to base vs finetune comparisons.<n>We introduce Dedicated Feature Crosscoders (DFCs), an architectural modification designed to better isolate features unique to one model.<n>Using this technique, we find in an unsupervised fashion features including Chinese Communist Party alignment in Qwen3-8B and Deepseek-R1-0528-Qwen3-8B, American exceptionalism in Llama3.1-8B-Instruct, and a copyright refusal mechanism in GPT-OSS-20B.
  arXiv  Detail & Related papers  (2026-02-12T08:53:25Z)
• 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)
• Gated Attention for Large Language Models: Non-linearity, Sparsity, and Attention-Sink-Free [81.65559031466452]
  We conduct experiments to investigate gating-augmented softmax attention variants.<n>We find that a simple modification-applying a head-specific sigmoid gate after the Scaled Dot-Product Attention (SDPA)-consistently improves performance.
  arXiv  Detail & Related papers  (2025-05-10T17:15:49Z)
• GenEDA: Towards Generative Netlist Functional Reasoning via Cross-Modal Circuit Encoder-Decoder Alignment [8.115489346573918]
  Existing pre-trained circuit foundation models are typically limited to standalone encoders for predictive tasks or decoders for generative tasks.<n>We present GenEDA, the first framework that cross-modally aligns circuit encoders with decoders within a shared latent space.<n>We leverage this architecture to develop the first generative foundation model for netlists.
  arXiv  Detail & Related papers  (2025-04-13T08:56:22Z)
• FTP: A Fine-grained Token-wise Pruner for Large Language Models via Token Routing [17.01412432658081]
  Large language models (LLMs) have demonstrated superior performance across various tasks by adhering to scaling laws.<n>We propose a fine-grained token-wise pruning approach for the LLMs, which presents a learnable router to adaptively identify the less important tokens.<n>Our approach achieves state-of-the-art (SOTA) pruning results, surpassing other existing pruning methods.
  arXiv  Detail & Related papers  (2024-12-16T07:09:46Z)
• Have Faith in Faithfulness: Going Beyond Circuit Overlap When Finding Model Mechanisms [35.514624827207136]
  Edge attribution patching (EAP), gradient-based approximation to interventions, has emerged as a scalable but imperfect solution to this problem. We introduce a new method - EAP with integrated gradients (EAP-IG) - that aims to better maintain a core property of circuits: faithfulness. Our experiments demonstrate that circuits found using EAP are less faithful than those found using EAP-IG, even though both have high node overlap with circuits found previously using causal interventions.
  arXiv  Detail & Related papers  (2024-03-26T15:44:58Z)
• Enhancing Multiple Reliability Measures via Nuisance-extended Information Bottleneck [77.37409441129995]
  In practical scenarios where training data is limited, many predictive signals in the data can be rather from some biases in data acquisition. We consider an adversarial threat model under a mutual information constraint to cover a wider class of perturbations in training. We propose an autoencoder-based training to implement the objective, as well as practical encoder designs to facilitate the proposed hybrid discriminative-generative training.
  arXiv  Detail & Related papers  (2023-03-24T16:03:21Z)
• Learning Target-aware Representation for Visual Tracking via Informative Interactions [49.552877881662475]
  We introduce a novel backbone architecture to improve target-perception ability of feature representation for tracking. The proposed GIM module and InBN mechanism are general and applicable to different backbone types including CNN and Transformer.
  arXiv  Detail & Related papers  (2022-01-07T16:22:27Z)
• DAIS: Automatic Channel Pruning via Differentiable Annealing Indicator Search [55.164053971213576]
  convolutional neural network has achieved great success in fulfilling computer vision tasks despite large computation overhead. Structured (channel) pruning is usually applied to reduce the model redundancy while preserving the network structure. Existing structured pruning methods require hand-crafted rules which may lead to tremendous pruning space.
  arXiv  Detail & Related papers  (2020-11-04T07:43:01Z)

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.