Scaling Interpretable Transformers with Parity Bottleneck Layers

• URL: http://arxiv.org/abs/2607.20652v1
• Date: Wed, 22 Jul 2026 18:25:16 GMT
• Title: Scaling Interpretable Transformers with Parity Bottleneck Layers
• Abstract summary: We introduce a GPT-2-scale architecture whose intermediate representations are efficient and wide / sparse by design.<n>A Deep Parity Bottleneck (DPB) replaces a learned over-complete basis with a parameter-free algebraic dictionary.<n>The ParityTransformer's features are native to the model's forwards pass by construction, addressing the question of whether SAEs probe features the model actually uses during computation.
• Score: 0.0
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Language models are thought to exhibit the phenomenon of superposition, representing many more features than dimensions in their residual streams. Sparse autoencoders (SAEs) are designed to recover such features post-hoc, but training models that are interpretable by construction has remained impractical, as a per-layer over-complete bottleneck is prohibitively expensive in both memory and compute. To overcome this issue, we introduce the ParityTransformer, a GPT-2-scale architecture whose intermediate representations are efficient and wide / sparse by design. At each layer, a Deep Parity Bottleneck (DPB) replaces a learned over-complete basis with a parameter-free algebraic dictionary, providing a deterministic incoherence guarantee and eliminating the memory requirements that have prevented per-layer interpretable bottlenecks at scale. A DPB is a hierarchically structured sparse bottleneck which efficiently enforces sparsity using a multi-level mixture-of-experts approach: a hardware-aware implementation that closes the cost gap between activation sparse and dense training to a manageable interpretability tax. Empirically, ParityTransformers perform at least as well as post-hoc SAEs on sparse probing tasks, while out-performing on measures of feature absorption, steering effectiveness, and fine-grained causal interventions. Because subsequent computation acts only on features that survive the sparse bottleneck, the ParityTransformer's features are native to the model's forwards pass by construction, addressing the question of whether SAEs probe features the model actually uses during computation. We see this as a step toward training models whose internal representations are interpretable by design rather than recovered post hoc.

Related papers

• Adaptive Depth Sparse Framework: Similarity-Driven Resource Allocation for Pre-Trained LLMs [18.457250148137494]
  We present an Adaptive Depth Sparse Framework (AdaDSF) that converts off-the-shelf pre-trained large language models into depth-sparse models without retraining.<n>On GPT-NeoX and Qwen2.5 over language modeling and commonsense reasoning, AdaDSF substantially reduces inference FLOPs while preserving performance close to dense counterparts.
  arXiv  Detail & Related papers  (2026-07-23T13:13:04Z)
• Spatial-Aware Reduction Framework: Towards Efficient and Faithful Visual State Space Models [71.59431204970339]
  STORM is a spatial-aware token reduction framework.<n>It reformulates reduction into a structured operation on spatial units.<n>It achieves state-of-the-art pruning accuracy across diverse vision Mamba backbones.
  arXiv  Detail & Related papers  (2026-06-18T08:31:11Z)
• Variable-Width Transformers [57.07167443557886]
  We investigate nonuniform capacity allocation across network depth by proposing a $times$-shaped > former architecture.<n>Our design maintains wider early and late layers while narrowing the middle layers, utilizing a parameter-free residual resizing mechanism.<n>By reducing the average layer width, this architecture also requires fewer overall FLOPs and smaller KV cache memory and I/O cost.
  arXiv  Detail & Related papers  (2026-06-16T17:59:03Z)
• The Expressivity Boundary of Probabilistic Circuits: A Comparison with Large Language Models [49.67598869075125]
  In autoregressive language modeling, Probabilistic Circuits (PCs) lag behind Transformer-based large language models (LLMs)<n>We show that structured-decomposable PCs can match Transformer separation rank on vtree-aligned partitions, but that this capacity is limited to partitions aligned with the fixed routing structure.<n>We further prove that decomposable PCs are strictly more expressive than structured-decomposable ones, though effectively optimizing them remains an open challenge.
  arXiv  Detail & Related papers  (2026-05-13T03:22:10Z)
• Latent-Condensed Transformer for Efficient Long Context Modeling [60.72493959155964]
  We propose Latent-Condensed Attention (LCA) for large language models.<n>LCA condenses context within MLA's latent space, where representation is disentangled into semantic latent vectors and positional keys.<n>LCA achieves up to 2.5$times$ prefilling speedup and 90% KV cache reduction at 128K context.
  arXiv  Detail & Related papers  (2026-04-14T08:40:31Z)
• Dual Path Attribution: Efficient Attribution for SwiGLU-Transformers through Layer-Wise Target Propagation [5.186807923082922]
  We introduce Dual Path Attribution (DPA), a novel framework that faithfully traces information flow on the frozen transformer in one forward and one backward pass.<n>DPA achieves state-of-the-art faithfulness and unprecedented efficiency compared to existing baselines.
  arXiv  Detail & Related papers  (2026-03-20T08:28:24Z)
• PT$^2$-LLM: Post-Training Ternarization for Large Language Models [52.4629647715623]
  Large Language Models (LLMs) have shown impressive capabilities across diverse tasks, but their large memory and compute demands hinder deployment.<n>We propose PT$2$-LLM, a post-training ternarization framework tailored for LLMs.<n>At its core is an Asymmetric Ternary Quantizer equipped with a two-stage refinement pipeline.
  arXiv  Detail & Related papers  (2025-09-27T03:01:48Z)
• Echo State Transformer: Attention Over Finite Memories [2.118933003468525]
  We introduce Echo State Transformers (EST), a hybrid architecture that elegantly resolves the challenge of sequential data processing.<n>EST integrates the Transformer attention mechanisms with principles from Reservoir Computing to create a fixed-size window distributed memory system.<n>EST ranks first overall in two of five categories, outperforming strong state-of-the-art baselines on classification and anomaly detection tasks.
  arXiv  Detail & Related papers  (2025-06-25T09:56:25Z)
• FLAT-LLM: Fine-grained Low-rank Activation Space Transformation for Large Language Model Compression [15.784158079414235]
  FLAT-LLM is a training-free structural compression method based on fine-grained low-rank transformations in the activation space.<n>It achieves efficient and effective weight compression without recovery fine-tuning, which could complete the calibration within a few minutes.
  arXiv  Detail & Related papers  (2025-05-29T19:42:35Z)
• Dynamic Context Pruning for Efficient and Interpretable Autoregressive Transformers [29.319666323947708]
  We present a novel approach that dynamically prunes contextual information while preserving the model's expressiveness. Our method employs a learnable mechanism that determines which uninformative tokens can be dropped from the context. Our reference implementation achieves up to $2times$ increase in inference throughput and even greater memory savings.
  arXiv  Detail & Related papers  (2023-05-25T07:39:41Z)
• PnP-DETR: Towards Efficient Visual Analysis with Transformers [146.55679348493587]
  Recently, DETR pioneered the solution vision tasks with transformers, it directly translates the image feature map into the object result. Recent transformer-based image recognition model andTT show consistent efficiency gain.
  arXiv  Detail & Related papers  (2021-09-15T01:10:30Z)

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.