Reroute turns irreversible visual-token pruning into recoverable routing that reuses existing attention scores, improving grounding performance under aggressive reduction on LLaVA-1.5 and Qwen while preserving TFLOPs and KV-cache budgets.
arXiv preprint arXiv:2506.21103 , year=
6 Pith papers cite this work. Polarity classification is still indexing.
6
Pith papers citing it
abstract
Conditional computation is a popular strategy to make Transformers more efficient. Existing methods often target individual modules (e.g., mixture-of-experts layers) or skip layers independently of one another. However, interpretability research has demonstrated that the middle layers of Transformers exhibit greater redundancy, and that early layers aggregate information into token positions. Guided by these insights, we propose a novel architecture that dynamically skips a variable number of layers from the middle outward. In particular, a learned gating mechanism determines whether to bypass a symmetric span of central blocks based on the input, and a gated attention mechanism prevents subsequent tokens from attending to skipped token positions. Residual norms are controlled with a 'sandwich' or 'perilayernorm' scheme and gate sparsity with an adaptive regularization loss. We had aimed to reduce compute requirements for 'simpler' tokens and potentially foster an emergent multi-level representational hierarchy but, at the scales investigated, our approach does not achieve improvements in the trade-off between validation cross-entropy and estimated FLOPs compared to dense baselines with fewer layers. We release our code at https://github.com/tim-lawson/skip-middle.
citation-role summary
background 1
citation-polarity summary
years
2026 6roles
background 1polarities
unclear 1representative citing papers
Joint few-step distillation and step-specific structural pruning turns a video diffusion model into a compact Mixture-of-Models that cuts 24% extra FLOPs per step and reaches 30× speedup on Wan-14B.
BEAM uses binary expert activation masks trained end-to-end to achieve dynamic sparsity in MoE models, cutting FLOPs by 85% with over 98% performance retention.
Task-aware pruning improves OOD model performance by realigning distorted OOD layerwise norm and pairwise-distance profiles with the task-adapted geometry observed on ID inputs.
BudgetFormer dynamically allocates a variable number of attention heads per input via a learned budget and relevance scoring, reducing inference cost on text classification while maintaining accuracy.
The survey organizes mechanistic interpretability techniques into a Locate-Steer-Improve framework to enable actionable improvements in LLM alignment, capability, and efficiency.
citing papers explorer
-
Reroute, Don't Remove: Recoverable Visual Token Routing for Vision-Language Models
Reroute turns irreversible visual-token pruning into recoverable routing that reuses existing attention scores, improving grounding performance under aggressive reduction on LLaVA-1.5 and Qwen while preserving TFLOPs and KV-cache budgets.
-
Dynamic-in-Few-Step: Unifying Dynamic Computation and Few-Step Distillation for Efficient Video Generation
Joint few-step distillation and step-specific structural pruning turns a video diffusion model into a compact Mixture-of-Models that cuts 24% extra FLOPs per step and reaches 30× speedup on Wan-14B.
-
BEAM: Binary Expert Activation Masking for Dynamic Routing in MoE
BEAM uses binary expert activation masks trained end-to-end to achieve dynamic sparsity in MoE models, cutting FLOPs by 85% with over 98% performance retention.
-
TAPIOCA: Why Task- Aware Pruning Improves OOD model Capability
Task-aware pruning improves OOD model performance by realigning distorted OOD layerwise norm and pairwise-distance profiles with the task-adapted geometry observed on ID inputs.
-
Adaptive Head Budgeting for Efficient Multi-Head Attention
BudgetFormer dynamically allocates a variable number of attention heads per input via a learned budget and relevance scoring, reducing inference cost on text classification while maintaining accuracy.
-
Locate, Steer, and Improve: A Practical Survey of Actionable Mechanistic Interpretability in Large Language Models
The survey organizes mechanistic interpretability techniques into a Locate-Steer-Improve framework to enable actionable improvements in LLM alignment, capability, and efficiency.