A hierarchical anisometric positional encoding that combines distance-indexed in-route and depot-anchored angular cross-route components improves transformer-based solvers for vehicle routing problems over index-based alternatives.
hub Canonical reference
Train Short, Test Long: Attention with Linear Biases Enables Input Length Extrapolation
Canonical reference. 71% of citing Pith papers cite this work as background.
53
Pith papers citing it
Background
71% of classified citations
abstract
Since the introduction of the transformer model by Vaswani et al. (2017), a fundamental question has yet to be answered: how does a model achieve extrapolation at inference time for sequences that are longer than it saw during training? We first show that extrapolation can be enabled by simply changing the position representation method, though we find that current methods do not allow for efficient extrapolation. We therefore introduce a simpler and more efficient position method, Attention with Linear Biases (ALiBi). ALiBi does not add positional embeddings to word embeddings; instead, it biases query-key attention scores with a penalty that is proportional to their distance. We show that this method trains a 1.3 billion parameter model on input sequences of length 1024 that extrapolates to input sequences of length 2048, achieving the same perplexity as a sinusoidal position embedding model trained on inputs of length 2048 but training 11% faster and using 11% less memory. ALiBi's inductive bias towards recency also leads it to outperform multiple strong position methods on the WikiText-103 benchmark.
hub tools
citation-role summary
background 4
method 2
other 1
citation-polarity summary
representative citing papers
ALiBi bias is the expectation of positional LSH-induced block masks, yielding spectral and max-norm approximation bounds that reduce long-context biased attention to randomized short-context unbiased attention.
Stealth Pretraining Seeding plants persistent unsafe behaviors in LLMs via diffuse poisoned web content that activates on precise triggers and evades standard evaluation.
URoPE is a parameter-free relative position embedding for transformers that works across arbitrary geometric spaces by ray sampling and projection, yielding consistent gains on novel view synthesis, 3D detection, tracking, and depth estimation.
Multi-head Gaussian kernels inject temporal scale discrepancy as inductive bias to enable full-duplex talking-listening avatar generation, supported by a new decoupled VoxHear dataset and claimed SOTA naturalness.
GRAPE unifies RoPE and ALiBi as special cases of group actions on positions, providing a principled design space for positional encodings via SO(d) rotations and GL unipotent transformations.
Transformers with O(sum m^j) blocks and O(d sum m^j) parameters can exactly interpolate any finite dataset of input sequences in R^d to output sequences of lengths m^j.
Infini-attention combines compressive memory with masked local attention and long-term linear attention inside each Transformer block to support infinite context length with bounded resources.
Massive activations are constant large values in LLMs that function as indispensable bias terms and concentrate attention probabilities on specific tokens.
Explicitly disentangling semantic and positional streams in a Transformer encoder reveals that absolute positional representations collapse to a 2D document-structure manifold, attention heads specialize by role, and the approach improves linguistic probing performance on 49 of 65 phenomena.
Self-pretraining improves Transformer sequence classification by enabling learning of proximity-biased attention from positional encodings that label supervision alone cannot easily acquire from random starts.
PMNet uses unitary phasor dynamics and hierarchical anchors to make explicit memory stable for long sequences, matching a 3x larger Mamba model on long-context robustness with a 119M parameter network.
Attention to goal tokens declines in multi-turn LLM interactions while residual representations often retain decodable goal information, and the gap between these predicts whether goal-conditioned behavior survives.
Theoretical analysis of continual factual knowledge acquisition shows data replay stabilizes pretrained knowledge by shifting convergence dynamics while regularization only slows forgetting, leading to the STOC method for attention-based replay selection.
EXACT re-allocates training supervision by inverse frequency of long effective-context targets, improving NoLiMa and RULER scores by 5-18 points on Qwen and LLaMA models without degrading standard QA or reasoning.
GAPE augments RoPE with query- and key-dependent gates to stabilize attention and improve long-context performance in language models.
HAGE proposes a trainable weighted graph memory framework with LLM intent classification, dynamic edge modulation, and RL optimization that improves long-horizon reasoning accuracy in agentic LLMs over static baselines.
FocuSFT uses an inner optimization loop to adapt fast-weight parameters into a parametric memory that sharpens attention on relevant content, then conditions outer-loop supervised fine-tuning on this representation, yielding gains on long-context benchmarks.
A mean-pool deep set trained on sets of size at most two produces an encoder that generalizes to arbitrary sizes, decoupling representation learning from posterior modeling and making training cost independent of deployment set size N.
ADE scales multi-anchor word representations to transformers via Vocabulary Projection, Grouped Positional Encoding, and context-aware reweighting, achieving 98.7% fewer trainable parameters than DeBERTa-v3-base while matching or exceeding it on two text-classification benchmarks and compressing the
Recurrent Transformers add per-layer recurrent memory via self-attention on own activations plus a tiling algorithm that reduces training memory traffic, yielding better C4 pretraining cross-entropy than parameter-matched standard transformers with fewer layers.
LoopGuard detects attention collapse loops during LLM decoding and prunes repetitive KV cache tail spans under fixed budget, cutting loop incidence by over 90 percentage points on the new LoopBench benchmark.
SharedLLM stacks two copies of a short-context LLM so the lower one compresses context into query-aware multi-grained tokens that are injected only at the lowest layers of the upper one, enabling generalization from 8K training to 128K+ inputs.
Flashlight is a compiler-native PyTorch framework that generates efficient fused kernels for arbitrary and data-dependent attention variants, supporting more cases than FlexAttention with competitive performance.
citing papers explorer
-
Rethinking Positional Encoding for Neural Vehicle Routing
A hierarchical anisometric positional encoding that combines distance-indexed in-route and depot-anchored angular cross-route components improves transformer-based solvers for vehicle routing problems over index-based alternatives.
-
Positional LSH: Binary Block Matrix Approximation for Attention with Linear Biases
ALiBi bias is the expectation of positional LSH-induced block masks, yielding spectral and max-norm approximation bounds that reduce long-context biased attention to randomized short-context unbiased attention.
-
PermaFrost-Attack: Stealth Pretraining Seeding(SPS) for planting Logic Landmines During LLM Training
Stealth Pretraining Seeding plants persistent unsafe behaviors in LLMs via diffuse poisoned web content that activates on precise triggers and evades standard evaluation.
-
URoPE: Universal Relative Position Embedding across Geometric Spaces
URoPE is a parameter-free relative position embedding for transformers that works across arbitrary geometric spaces by ray sampling and projection, yielding consistent gains on novel view synthesis, 3D detection, tracking, and depth estimation.
-
Beyond Monologue: Interactive Talking-Listening Avatar Generation with Conversational Audio Context-Aware Kernels
Multi-head Gaussian kernels inject temporal scale discrepancy as inductive bias to enable full-duplex talking-listening avatar generation, supported by a new decoupled VoxHear dataset and claimed SOTA naturalness.
-
Group Representational Position Encoding
GRAPE unifies RoPE and ALiBi as special cases of group actions on positions, providing a principled design space for positional encodings via SO(d) rotations and GL unipotent transformations.
-
Exact Sequence Interpolation with Transformers
Transformers with O(sum m^j) blocks and O(d sum m^j) parameters can exactly interpolate any finite dataset of input sequences in R^d to output sequences of lengths m^j.
-
Leave No Context Behind: Efficient Infinite Context Transformers with Infini-attention
Infini-attention combines compressive memory with masked local attention and long-term linear attention inside each Transformer block to support infinite context length with bounded resources.
-
Massive Activations in Large Language Models
Massive activations are constant large values in LLMs that function as indispensable bias terms and concentrate attention probabilities on specific tokens.
-
Give it Space! Explicit Disentangling of Positional and Semantic Representations in Encoders
Explicitly disentangling semantic and positional streams in a Transformer encoder reveals that absolute positional representations collapse to a 2D document-structure manifold, attention heads specialize by role, and the approach improves linguistic probing performance on 49 of 65 phenomena.
-
Towards Understanding Self-Pretraining for Sequence Classification
Self-pretraining improves Transformer sequence classification by enabling learning of proximity-biased attention from positional encodings that label supervision alone cannot easily acquire from random starts.
-
Phasor Memory Networks: Stable Backpropagation Through Time for Scalable Explicit Memory
PMNet uses unitary phasor dynamics and hierarchical anchors to make explicit memory stable for long sequences, matching a 3x larger Mamba model on long-context robustness with a 119M parameter network.
-
When Attention Closes: How LLMs Lose the Thread in Multi-Turn Interaction
Attention to goal tokens declines in multi-turn LLM interactions while residual representations often retain decodable goal information, and the gap between these predicts whether goal-conditioned behavior survives.
-
Towards Understanding Continual Factual Knowledge Acquisition of Language Models: From Theory to Algorithm
Theoretical analysis of continual factual knowledge acquisition shows data replay stabilizes pretrained knowledge by shifting convergence dynamics while regularization only slows forgetting, leading to the STOC method for attention-based replay selection.
-
Where Does Long-Context Supervision Actually Go? Effective-Context Exposure Balancing
EXACT re-allocates training supervision by inverse frequency of long effective-context targets, improving NoLiMa and RULER scores by 5-18 points on Qwen and LLaMA models without degrading standard QA or reasoning.
-
Remember to Forget: Gated Adaptive Positional Encoding
GAPE augments RoPE with query- and key-dependent gates to stabilize attention and improve long-context performance in language models.
-
HAGE: Harnessing Agentic Memory via RL-Driven Weighted Graph Evolution
HAGE proposes a trainable weighted graph memory framework with LLM intent classification, dynamic edge modulation, and RL optimization that improves long-horizon reasoning accuracy in agentic LLMs over static baselines.
-
FocuSFT: Bilevel Optimization for Dilution-Aware Long-Context Fine-Tuning
FocuSFT uses an inner optimization loop to adapt fast-weight parameters into a parametric memory that sharpens attention on relevant content, then conditions outer-loop supervised fine-tuning on this representation, yielding gains on long-context benchmarks.
-
It Just Takes Two: Scaling Amortized Inference to Large Sets
A mean-pool deep set trained on sets of size at most two produces an encoder that generalizes to arbitrary sizes, decoupling representation learning from posterior modeling and making training cost independent of deployment set size N.
-
ADE: Adaptive Dictionary Embeddings -- Scaling Multi-Anchor Representations to Large Language Models
ADE scales multi-anchor word representations to transformers via Vocabulary Projection, Grouped Positional Encoding, and context-aware reweighting, achieving 98.7% fewer trainable parameters than DeBERTa-v3-base while matching or exceeding it on two text-classification benchmarks and compressing the
-
The Recurrent Transformer: Greater Effective Depth and Efficient Decoding
Recurrent Transformers add per-layer recurrent memory via self-attention on own activations plus a tiling algorithm that reduces training memory traffic, yielding better C4 pretraining cross-entropy than parameter-matched standard transformers with fewer layers.
-
LoopGuard: Breaking Self-Reinforcing Attention Loops via Dynamic KV Cache Intervention
LoopGuard detects attention collapse loops during LLM decoding and prunes repetitive KV cache tail spans under fixed budget, cutting loop incidence by over 90 percentage points on the new LoopBench benchmark.
-
Stacked from One: Multi-Scale Self-Injection for Context Window Extension
SharedLLM stacks two copies of a short-context LLM so the lower one compresses context into query-aware multi-grained tokens that are injected only at the lowest layers of the upper one, enabling generalization from 8K training to 128K+ inputs.
-
Flashlight: PyTorch Compiler Extensions to Accelerate Attention Variants
Flashlight is a compiler-native PyTorch framework that generates efficient fused kernels for arbitrary and data-dependent attention variants, supporting more cases than FlexAttention with competitive performance.
-
Mixture-of-Experts Can Surpass Dense LLMs Under Strictly Equal Resource
MoE models with activation rates in an optimal region outperform dense LLMs of identical total parameter count, training compute, and data budget, with the optimal region consistent across scales.
-
Long-Context Autoregressive Video Modeling with Next-Frame Prediction
FAR baseline plus asymmetric kernels for long short-term context modeling achieves SOTA short and long video generation in autoregressive setups.
-
Flex Attention: A Programming Model for Generating Optimized Attention Kernels
FlexAttention supplies a compiler-driven interface that expresses common attention variants in a few lines of PyTorch and emits optimized kernels whose speed matches hand-written implementations.
-
Condense, Don't Just Prune: Enhancing Efficiency and Performance in MoE Layer Pruning
CD-MoE condenses fine-grained MoE layers with shared experts into dense layers, retaining 90% accuracy with 27.5% memory cut and 1.26x speedup on DeepSeekMoE-16B, recovering 98% via brief fine-tuning.
-
When Attention Sink Emerges in Language Models: An Empirical View
Attention sinks emerge in language models from softmax-induced token dependence on attention scores and do not appear when using sigmoid attention without normalization in models up to 1B parameters.
-
Gated Linear Attention Transformers with Hardware-Efficient Training
Gated linear attention Transformers achieve competitive language modeling results with linear-time inference, superior length generalization, and higher training throughput than Mamba.
-
MemGPT: Towards LLMs as Operating Systems
MemGPT uses OS-inspired virtual context management to extend LLM context windows for large document analysis and long-term multi-session chat.
-
A Measure-Theoretic Analysis of Reasoning: Structural Generalization and Approximation Limits
Applies optimal transport to bound OOD generalization error in Transformers via Lipschitz continuity and TC^0 circuit depth lower bounds for Dyck-k backtracking, supported by evaluations on 54 configurations.
-
Supporting System Testing with a Multi-Agent LLM-based Framework for Knowledge Graph Extraction: A Case Study with Ethernet Switch Systems
A multi-agent LLM-based framework extracts knowledge graphs from 50 real Ethernet switch manuals with 0.97-0.99 correctness to enable downstream test case specification generation.
-
Kaczmarz Linear Attention
Kaczmarz Linear Attention replaces the empirical coefficient in Gated DeltaNet with a key-norm-normalized step size derived from the online regression objective, yielding lower perplexity and better needle-in-haystack performance.
-
FAAST: Forward-Only Associative Learning via Closed-Form Fast Weights for Test-Time Supervised Adaptation
FAAST performs test-time supervised adaptation by analytically deriving fast weights from examples in one forward pass, matching backprop performance with over 90% less adaptation time and up to 95% memory savings versus memory-based methods.
-
Decouple and Cache: KV Cache Construction for Streaming Video Understanding
DSCache decouples cumulative past and instant KV caches with position-agnostic encoding to adapt offline VideoVLLMs to streaming video, delivering 2.5% average accuracy gains on QA benchmarks.
-
Adaptive 3D-RoPE: Physics-Aligned Rotary Positional Encoding for Wireless Foundation Models
Adaptive 3D-RoPE adapts rotary positional encoding to wireless channel physics via learnable 3D frequencies and dynamic CSI control, yielding up to 10.7 dB NMSE gains in scale extrapolation and 1 dB in zero-shot tasks.
-
Working Memory Constraints Scaffold Learning in Transformers under Data Scarcity
Fixed-width and decay-based attention mechanisms inspired by working memory improve Transformer grammatical accuracy and human alignment under limited training data.
-
Learning Class Difficulty in Imbalanced Histopathology Segmentation via Dynamic Focal Attention
Dynamic Focal Attention learns class-specific difficulty via per-class biases in attention logits, improving Dice and IoU on imbalanced histopathology segmentation benchmarks.
-
Voxtral TTS
Voxtral TTS produces expressive multilingual speech from 3-second reference audio with a hybrid autoregressive-plus-flow-matching architecture and a new VQ-FSQ tokenizer, achieving 68.4% win rate over ElevenLabs in human evaluations.
-
Video Parallel Scaling: Aggregating Diverse Frame Subsets for VideoLLMs
Video Parallel Scaling improves VideoLLM performance by aggregating outputs from parallel inferences on complementary disjoint frame subsets, effectively contracting the Chinchilla scaling law via uncorrelated visual evidence.
-
F5-TTS: A Fairytaler that Fakes Fluent and Faithful Speech with Flow Matching
F5-TTS generates natural speech from text via flow matching on DiT with simple text padding, ConvNeXt refinement, and sway sampling, trained on 100K hours multilingual data.
-
Galactica: A Large Language Model for Science
Galactica, a science-specialized LLM, reports higher scores than GPT-3, Chinchilla, and PaLM on LaTeX knowledge, mathematical reasoning, and medical QA benchmarks while outperforming general models on BIG-bench.
-
Simple Token-Efficient Vision-Language Model for Case-level Pathology Synoptic Report Generation
A token-efficient VLM with frozen encoder, two-layer MLP aligner, and LLM decoder generates case-level synoptic pathology reports from multi-WSI inputs using 5x magnification patches and two-stage supervised training.
-
Hybrid Architectures for Language Models: Systematic Analysis and Design Insights
This work systematically compares inter-layer and intra-layer hybridization strategies for combining self-attention and Mamba-style state space models, evaluating them on language modeling, downstream tasks, long-context performance, scaling, and efficiency to derive optimal design recipes.
-
Step-Video-T2V Technical Report: The Practice, Challenges, and Future of Video Foundation Model
Step-Video-T2V describes a 30B-parameter text-to-video model with custom Video-VAE, 3D DiT, flow matching, and Video-DPO that claims state-of-the-art results on a new internal benchmark.
-
Baichuan 2: Open Large-scale Language Models
Baichuan 2 presents 7B and 13B LLMs trained on 2.6T tokens that match or exceed similar open models on MMLU, CMMLU, GSM8K, HumanEval and excel in medicine and law.
-
Positional Encoding in Transformer-Based Time Series Models: A Survey
A survey of positional encoding methods in transformer-based time series models that evaluates fixed, learnable, relative, and hybrid approaches on classification tasks and links effectiveness to data characteristics.
-
A Survey on Large Language Models for Code Generation
A systematic literature review that organizes recent work on LLMs for code generation into a taxonomy covering data curation, model advances, evaluations, ethics, environmental impact, and applications, with benchmark comparisons.
-
Large Language Models: A Survey
The paper surveys key large language models, their training methods, datasets, evaluation benchmarks, and future research directions in the field.