Shared Global KV with Layer-Specific Local History

arXiv:2609.28006 · cs.LG · Submitted 2026-09-23 · Read on arXiv

cs.LG

Submitted: 2026-09-23

Updated: 2026-09-23

Code: https://github.com/allenai/longformer

Terminology

• Perplexity — 22× in this paper · explained in 3 episodes
• Adaptation — 21× in this paper · explained in 1 episode
• Decode — 19× in this paper · explained in 2 episodes
• Transformer — 15× in this paper · explained in 2 episodes
• Accuracy — 10× in this paper · explained in 1 episode
• Aggregation — 5× in this paper · explained in 1 episode
• Benchmark — 5× in this paper · explained in 2 episodes
• Allocator — 4× in this paper · explained in 1 episode
• Downstream Tasks — 3× in this paper · explained in 2 episodes
• KV cache — 2× in this paper · explained in 3 episodes
• Parallel Loop Transformer — 2× in this paper
• Pruning — 2× in this paper · explained in 4 episodes
• Adam — 1× in this paper · explained in 2 episodes
• Core — 1× in this paper · explained in 2 episodes
• Embeddings — 1× in this paper · explained in 2 episodes
• Generalization — 1× in this paper · explained in 13 episodes
• Provenance — 1× in this paper · explained in 4 episodes
• Regularization — 1× in this paper · explained in 3 episodes
• Sparse Attention — 1× in this paper · explained in 5 episodes
• Sparsity — 1× in this paper · explained in 4 episodes
• Transformer Models — 1× in this paper · explained in 2 episodes
• Verification — 1× in this paper · explained in 2 episodes

Sources

• Longformer: The Long-Document Transformer
• Reducing Transformer Key-Value Cache Size with Cross-Layer Attention
• Extending Context Window of Large Language Models via Positional Interpolation
• Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challenge
• FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awareness
• UniPrefill: Universal Long-Context Prefill Acceleration via Block-wise Dynamic Sparsification
• GoldFinch: High Performance RWKV/Transformer Hybrid with Linear Pre-Fill and Extreme KV-Cache Compression
• POP: Prefill-Only Pruning for Efficient Large Model Inference
• Block Transformer: Global-to-Local Language Modeling for Fast Inference
• Attention Residuals
• Decoupled Weight Decay Regularization
• The FineWeb Datasets: Decanting the Web for the Finest Text Data at Scale
• Compressive Transformers for Long-Range Sequence Modelling
• Inference-Friendly Models With MixAttention
• Fast Transformer Decoding: One Write-Head is All You Need
• GLU Variants Improve Transformer
• RoFormer: Enhanced Transformer with Rotary Position Embedding
• You Only Index Once: Cross-Layer Sparse Attention with Shared Routing
• Parallel Loop Transformer for Efficient Test-Time Computation Scaling
• Root Mean Square Layer Normalization

Related papers

• Polynomial-Augmented Neural Networks (PANNs) with Weak Orthogonality Constraints for Enhanced Function and PDE Approximation
• AIRL-S: Unifying Reinforcement Learning and Search-Based Test-Time Scaling via Adversarial Inverse Reinforcement Learning
• Transformers as Bayesian In-Context Experimenters: Smoothness-Adaptive Efficient ATE Estimation
• Convergence issues in Relational Concept Analysis based on AOC-posets
• Beliefs Beyond Posteriors: Local-Consistency Optimisation for Bayesian Neural Networks
• Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks