Finding Transformer Circuits with Edge Pruning

• URL: http://arxiv.org/abs/2406.16778v2
• Date: Thu, 05 Dec 2024 15:38:11 GMT
• Title: Finding Transformer Circuits with Edge Pruning
• Authors: Adithya Bhaskar, Alexander Wettig, Dan Friedman, Danqi Chen,
• Abstract summary: We propose Edge Pruning as an effective and scalable solution to automated circuit discovery. Our method finds circuits in GPT-2 that use less than half the number of edges compared to circuits found by previous methods. Thanks to its efficiency, we scale Edge Pruning to CodeLlama-13B, a model over 100x the scale that prior methods operate on.
• Score: 71.12127707678961
• License:
• Abstract: The path to interpreting a language model often proceeds via analysis of circuits -- sparse computational subgraphs of the model that capture specific aspects of its behavior. Recent work has automated the task of discovering circuits. Yet, these methods have practical limitations, as they rely either on inefficient search algorithms or inaccurate approximations. In this paper, we frame automated circuit discovery as an optimization problem and propose *Edge Pruning* as an effective and scalable solution. Edge Pruning leverages gradient-based pruning techniques, but instead of removing neurons or components, it prunes the \emph{edges} between components. Our method finds circuits in GPT-2 that use less than half the number of edges compared to circuits found by previous methods while being equally faithful to the full model predictions on standard circuit-finding tasks. Edge Pruning is efficient even with as many as 100K examples, outperforming previous methods in speed and producing substantially better circuits. It also perfectly recovers the ground-truth circuits in two models compiled with Tracr. Thanks to its efficiency, we scale Edge Pruning to CodeLlama-13B, a model over 100x the scale that prior methods operate on. We use this setting for a case study comparing the mechanisms behind instruction prompting and in-context learning. We find two circuits with more than 99.96% sparsity that match the performance of the full model and reveal that the mechanisms in the two settings overlap substantially. Our case study shows that Edge Pruning is a practical and scalable tool for interpretability and sheds light on behaviors that only emerge in large models.

Related papers

• Position-aware Automatic Circuit Discovery [59.64762573617173]
  We identify a gap in existing circuit discovery methods, treating model components as equally relevant across input positions. We propose two improvements to incorporate positionality into circuits, even on tasks containing variable-length examples. Our approach enables fully automated discovery of position-sensitive circuits, yielding better trade-offs between circuit size and faithfulness compared to prior work.
  arXiv  Detail & Related papers  (2025-02-07T00:18:20Z)
• Pathwise optimization for bridge-type estimators and its applications [49.1574468325115]
  Pathwise methods allow to efficiently compute the full path for penalized estimators. We apply these algorithms to the penalized estimation of processes observed at discrete times.
  arXiv  Detail & Related papers  (2024-12-05T10:38:29Z)
• Transformer Circuit Faithfulness Metrics are not Robust [0.04260910081285213]
  We measure circuit 'faithfulness' by ablating portions of the model's computation. We conclude that existing circuit faithfulness scores reflect both the methodological choices of researchers as well as the actual components of the circuit. The ultimate goal of mechanistic interpretability work is to understand neural networks, so we emphasize the need for more clarity in the precise claims being made about circuits.
  arXiv  Detail & Related papers  (2024-07-11T17:59:00Z)
• CircuitVAE: Efficient and Scalable Latent Circuit Optimization [22.93567682576068]
  CircuitVAE is a search algorithm that embeds computation graphs in a continuous space. Our algorithm is highly sample-efficient, yet gracefully scales to large problem instances and high sample budgets. We find CircuitVAE can design state-of-the-art adders in a real-world chip, demonstrating that our method can outperform commercial tools in a realistic setting.
  arXiv  Detail & Related papers  (2024-06-13T18:47:52Z)
• Robust shallow shadows [0.251657752676152]
  We present a robust shadow estimation protocol for wide classes of shallow measurement circuits. We show how to estimate this directly from experimental data using tensor-network tools. Under the practical constraints of current and near-term noisy quantum devices, our method maximally realizes the potential of shadow estimation with global rotations.
  arXiv  Detail & Related papers  (2024-05-09T18:00:09Z)
• 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)
• Attribution Patching Outperforms Automated Circuit Discovery [3.8695554579762814]
  We show that a simple method based on attribution patching outperforms all existing methods. We apply a linear approximation to activation patching to estimate the importance of each edge in the computational subgraph.
  arXiv  Detail & Related papers  (2023-10-16T12:34:43Z)
• TransPath: Learning Heuristics For Grid-Based Pathfinding via Transformers [64.88759709443819]
  We suggest learning the instance-dependent proxies that are supposed to notably increase the efficiency of the search. The first proxy we suggest to learn is the correction factor, i.e. the ratio between the instance independent cost-to-go estimate and the perfect one. The second proxy is the path probability, which indicates how likely the grid cell is lying on the shortest path.
  arXiv  Detail & Related papers  (2022-12-22T14:26:11Z)
• CATRO: Channel Pruning via Class-Aware Trace Ratio Optimization [61.71504948770445]
  We propose a novel channel pruning method via Class-Aware Trace Ratio Optimization (CATRO) to reduce the computational burden and accelerate the model inference. We show that CATRO achieves higher accuracy with similar cost or lower cost with similar accuracy than other state-of-the-art channel pruning algorithms. Because of its class-aware property, CATRO is suitable to prune efficient networks adaptively for various classification subtasks, enhancing handy deployment and usage of deep networks in real-world applications.
  arXiv  Detail & Related papers  (2021-10-21T06:26:31Z)
• Manifold Regularized Dynamic Network Pruning [102.24146031250034]
  This paper proposes a new paradigm that dynamically removes redundant filters by embedding the manifold information of all instances into the space of pruned networks. The effectiveness of the proposed method is verified on several benchmarks, which shows better performance in terms of both accuracy and computational cost.
  arXiv  Detail & Related papers  (2021-03-10T03:59:03Z)

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.