paper-with-me

홈 › Papers

InterpBench: Semi-Synthetic Transformers for Evaluating Mechanistic Interpretability Techniques

2024-07-19 · Rohan Gupta, Iván Arcuschin, Thomas Kwa, Adrià Garriga-Alonso

Mechanistic interpretability methods aim to identify the algorithm a neural network implements, but it is difficult to validate such methods when the true algorithm is unknown. This work presents InterpBench, a collection of semi-synthetic yet realistic transformers with known circuits for evaluating these techniques. We train simple neural networks using a stricter version of Interchange Intervention Training (IIT) which we call Strict IIT (SIIT). Like the original, SIIT trains neural networks by aligning their internal computation with a desired high-level causal model, but it also prevents non-circuit nodes from affecting the model's output. We evaluate SIIT on sparse transformers produced by the Tracr tool and find that SIIT models maintain Tracr's original circuit while being more realistic. SIIT can also train transformers with larger circuits, like Indirect Object Identification (IOI). Finally, we use our benchmark to evaluate existing circuit discovery techniques.

📄 PDF Abstract BibTeX arXiv:2407.14494

Code (3)

FlyingPumba/circuits-benchmark 공식 구현 jax
flyingpumba/interpbench 공식 구현 jax
aaronmueller/mib pytorch

Similar Papers 제목 키워드 기반

Can Language Model Agents be Helpful Circuit Explainers in Mechanistic Interpretability?

2026-06-23 · Ayan Antik Khan, Harsh Kohli, Yuekun Yao, Huan Sun 외 arxiv

Mechanistic interpretability has made substantial progress in automatically localizing circuits, but explaining what localized components do remains labor-intensive and difficult to standardize. In this work, we study wh…

Mechanistic evaluation of Transformers and state space models

2025-05-21 · Aryaman Arora, Neil Rathi, Nikil Roashan Selvam, Róbert Csórdas 외

State space models (SSMs) for language modelling promise an efficient and performant alternative to quadratic-attention Transformers, yet show variable performance on recalling basic information from the context. While p…

Language ModellingMambaState Space Models

A Mechanistic Analysis of a Transformer Trained on a Symbolic Multi-Step Reasoning Task

2024-02-19 · Jannik Brinkmann, Abhay Sheshadri, Victor Levoso, Paul Swoboda 외

Transformers demonstrate impressive performance on a range of reasoning benchmarks. To evaluate the degree to which these abilities are a result of actual reasoning, existing work has focused on developing sophisticated …

Dissecting Multimodal In-Context Learning: Modality Asymmetries and Circuit Dynamics in modern Transformers

2026-01-28 · Yiran Huang, Karsten Roth, Quentin Bouniot, Wenjia Xu 외 arxiv

Transformer-based multimodal large language models often exhibit in-context learning (ICL) abilities. Motivated by this phenomenon, we ask: how do transformers learn to associate information across modalities from in-con…

Emergent Analogical Reasoning in Transformers

2026-02-02 · Gouki Minegishi, Jingyuan Feng, Hiroki Furuta, Takeshi Kojima 외 arxiv

Analogy is a central faculty of human intelligence, enabling abstract patterns discovered in one domain to be applied to another. Despite its central role in cognition, the mechanisms by which Transformers acquire and im…