LongBench is the first bilingual multi-task benchmark for long context understanding in LLMs, containing 21 datasets in 6 categories with average lengths of 6711 words (English) and 13386 characters (Chinese).
hub Mixed citations
Extending Context Window of Large Language Models via Positional Interpolation
Mixed citation behavior. Most common role is background (65%).
59
Pith papers citing it
Background
65% of classified citations
abstract
We present Position Interpolation (PI) that extends the context window sizes of RoPE-based pretrained LLMs such as LLaMA models to up to 32768 with minimal fine-tuning (within 1000 steps), while demonstrating strong empirical results on various tasks that require long context, including passkey retrieval, language modeling, and long document summarization from LLaMA 7B to 65B. Meanwhile, the extended model by Position Interpolation preserve quality relatively well on tasks within its original context window. To achieve this goal, Position Interpolation linearly down-scales the input position indices to match the original context window size, rather than extrapolating beyond the trained context length which may lead to catastrophically high attention scores that completely ruin the self-attention mechanism. Our theoretical study shows that the upper bound of interpolation is at least $\sim 600 \times$ smaller than that of extrapolation, further demonstrating its stability. Models extended via Position Interpolation retain its original architecture and can reuse most pre-existing optimization and infrastructure.
hub tools
citation-role summary
background 14
method 4
baseline 2
citation-polarity summary
representative citing papers
Autoregressive transformers exhibit measurable cognitive fatigue during extended generation, quantified by the Fatigue Index that predicts degradation (AUROC 0.95) and repetition (rho 0.94).
Tensor Cache augments sliding-window attention with an eviction-fed outer-product associative memory and a training correction to improve long-context performance under bounded memory.
ExtraVAR enables resolution extrapolation in visual autoregressive models by stage-aware RoPE remapping and entropy-driven attention scaling, suppressing repetition and detail loss.
Jordan-RoPE realizes a distance-modulated phase basis via non-semisimple Jordan blocks, generating features such as d e^{iωd} for relative positional encoding.
Dual Triangle Attention achieves effective bidirectional attention with built-in positional inductive bias via dual triangular masks, outperforming standard bidirectional attention on position-sensitive tasks and showing strong masked language modeling results with or without positional embeddings.
Multiscreen replaces softmax attention with screening to provide absolute query-key relevance, resulting in models with 30% fewer parameters that maintain stable performance at long contexts.
SHARP applies a spectrum-aware dynamic RoPE scaling schedule that promotes resolution more strongly in early denoising stages and relaxes it later, outperforming static baselines on quality metrics for remote sensing images.
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.
Stacked causal self-attention combined with LayerNorm induces recency bias in Transformer decoders, reversing the earlier-token bias seen in attention alone.
KV cache compression causes task-dependent degradation in high-density reasoning due to disrupted CoT links; ShotKV mitigates this by preserving few-shot examples as indivisible semantic units through phase separation, delivering 9-18% accuracy gains and 11% latency reduction.
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.
LongRoPE extends LLM context windows to 2048k tokens via search for non-uniform positional interpolation, progressive fine-tuning from 256k, and short-context readjustment.
LPES uses per-layer scaling factors optimized by a genetic algorithm with Bézier curves to balance attention and improve long-context LLM performance by up to 11.2% on key-value retrieval.
SEGA adaptively scales RoPE attention components using spectral-energy guidance from the latent to improve structural coherence and fine details in high-resolution DiT synthesis.
Continued pre-training with balanced long-document VQA data extends a 7B LVLM to 128K context, improving long-document VQA by 7.1% and generalizing to 512K without further training.
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.
KV-Fold turns frozen transformers into stable long-context models by folding the KV cache across sequence chunks in repeated forward passes.
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.
SCoL trains LLMs via meta-reinforcement learning to generate layer-specific update instructions that improve knowledge acquisition and retention from context streams over standard baselines.
RouteHead trains a lightweight router to dynamically select optimal LLM attention heads per query for improved attention-based document re-ranking.
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.
MMHNet enables video-to-audio models trained on short clips to generalize and generate audio for videos over 5 minutes long.
citing papers explorer
-
LongBench: A Bilingual, Multitask Benchmark for Long Context Understanding
LongBench is the first bilingual multi-task benchmark for long context understanding in LLMs, containing 21 datasets in 6 categories with average lengths of 6711 words (English) and 13386 characters (Chinese).
-
Cognitive Fatigue in Autoregressive Transformers: Formalization and Measurement
Autoregressive transformers exhibit measurable cognitive fatigue during extended generation, quantified by the Fatigue Index that predicts degradation (AUROC 0.95) and repetition (rho 0.94).
-
Tensor Cache: Eviction-conditioned Associative Memory for Transformers
Tensor Cache augments sliding-window attention with an eviction-fed outer-product associative memory and a training correction to improve long-context performance under bounded memory.
-
ExtraVAR: Stage-Aware RoPE Remapping for Resolution Extrapolation in Visual Autoregressive Models
ExtraVAR enables resolution extrapolation in visual autoregressive models by stage-aware RoPE remapping and entropy-driven attention scaling, suppressing repetition and detail loss.
-
Jordan-RoPE: Non-Semisimple Relative Positional Encoding via Complex Jordan Blocks
Jordan-RoPE realizes a distance-modulated phase basis via non-semisimple Jordan blocks, generating features such as d e^{iωd} for relative positional encoding.
-
Dual Triangle Attention: Effective Bidirectional Attention Without Positional Embeddings
Dual Triangle Attention achieves effective bidirectional attention with built-in positional inductive bias via dual triangular masks, outperforming standard bidirectional attention on position-sensitive tasks and showing strong masked language modeling results with or without positional embeddings.
-
Screening Is Enough
Multiscreen replaces softmax attention with screening to provide absolute query-key relevance, resulting in models with 30% fewer parameters that maintain stable performance at long contexts.
-
SHARP: Spectrum-aware Highly-dynamic Adaptation for Resolution Promotion in Remote Sensing Synthesis
SHARP applies a spectrum-aware dynamic RoPE scaling schedule that promotes resolution more strongly in early denoising stages and relaxes it later, outperforming static baselines on quality metrics for remote sensing images.
-
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.
-
LayerNorm Induces Recency Bias in Transformer Decoders
Stacked causal self-attention combined with LayerNorm induces recency bias in Transformer decoders, reversing the earlier-token bias seen in attention alone.
-
Semantic Integrity Matters: Benchmarking and Preserving High-Density Reasoning in KV Cache Compression
KV cache compression causes task-dependent degradation in high-density reasoning due to disrupted CoT links; ShotKV mitigates this by preserving few-shot examples as indivisible semantic units through phase separation, delivering 9-18% accuracy gains and 11% latency reduction.
-
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.
-
LongRoPE: Extending LLM Context Window Beyond 2 Million Tokens
LongRoPE extends LLM context windows to 2048k tokens via search for non-uniform positional interpolation, progressive fine-tuning from 256k, and short-context readjustment.
-
Mitigating Position Bias in Transformers via Layer-Specific Positional Embedding Scaling
LPES uses per-layer scaling factors optimized by a genetic algorithm with Bézier curves to balance attention and improve long-context LLM performance by up to 11.2% on key-value retrieval.
-
SEGA: Spectral-Energy Guided Attention for Resolution Extrapolation in Diffusion Transformers
SEGA adaptively scales RoPE attention components using spectral-energy guidance from the latent to improve structural coherence and fine details in high-resolution DiT synthesis.
-
Training Long-Context Vision-Language Models Effectively with Generalization Beyond 128K Context
Continued pre-training with balanced long-document VQA data extends a 7B LVLM to 128K context, improving long-document VQA by 7.1% and generalizing to 512K without further training.
-
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.
-
KV-Fold: One-Step KV-Cache Recurrence for Long-Context Inference
KV-Fold turns frozen transformers into stable long-context models by folding the KV cache across sequence chunks in repeated forward passes.
-
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.
-
Self-Consolidating Language Models: Continual Knowledge Incorporation from Context
SCoL trains LLMs via meta-reinforcement learning to generate layer-specific update instructions that improve knowledge acquisition and retention from context streams over standard baselines.
-
Learning to Route Queries to Heads for Attention-based Re-ranking with Large Language Models
RouteHead trains a lightweight router to dynamically select optimal LLM attention heads per query for improved attention-based document re-ranking.
-
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.
-
Echoes Over Time: Unlocking Length Generalization in Video-to-Audio Generation Models
MMHNet enables video-to-audio models trained on short clips to generalize and generate audio for videos over 5 minutes long.
-
HyMem: Hybrid Memory Architecture with Dynamic Retrieval Scheduling
HyMem introduces dual-granular memory storage with a lightweight summary module for fast responses and selective activation of a deep LLM module for complex queries, outperforming full-context baselines by 92.6% lower computational cost on LOCOMO and LongMemEval benchmarks.
-
MTraining: Distributed Dynamic Sparse Attention for Efficient Ultra-Long Context Training
MTraining scales LLM training to 512K-token contexts on 32 A100 GPUs by integrating dynamic sparse training patterns with balanced and hierarchical sparse ring attention, achieving up to 6x throughput gains without accuracy loss on long-context benchmarks.
-
Self-Forcing++: Towards Minute-Scale High-Quality Video Generation
Self-Forcing++ scales autoregressive video diffusion to over 4 minutes by using self-generated segments for guidance, reducing error accumulation and outperforming baselines in fidelity and consistency.
-
Positional Encoding via Token-Aware Phase Attention
TAPA adds a learnable phase function to attention to preserve long-range token interactions, enabling direct continual pretraining, length extrapolation, lower perplexity, and stronger retrieval than RoPE-style methods.
-
MemAgent: Reshaping Long-Context LLM with Multi-Conv RL-based Memory Agent
MemAgent uses multi-conversation RL to train a memory agent that reads text in segments and overwrites memory, extrapolating from 8K training to 3.5M token QA with under 5% loss and 95%+ on 512K RULER.
-
Gated Attention for Large Language Models: Non-linearity, Sparsity, and Attention-Sink-Free
Applying a head-specific sigmoid gate after SDPA in LLMs boosts performance and stability by adding non-linearity and query-dependent sparse modulation while reducing attention sinks.
-
MegaScale-Data: Scaling Dataloader for Multisource Large Foundation Model Training
MegaScale-Data is a distributed data loading system that disaggregates preprocessing and applies auto-partitioning to deliver 4.5x higher end-to-end training throughput and 13.5x lower CPU memory usage for multisource large foundation models.
-
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.
-
TabICL: A Tabular Foundation Model for In-Context Learning on Large Data
TabICL scales in-context learning to large tabular data via column-then-row attention for row embeddings followed by a transformer, matching TabPFNv2 speed and performance while outperforming it and CatBoost on datasets over 10K samples.
-
Long Context Transfer from Language to Vision
Extending language model context length enables LMMs to process over 200K visual tokens from long videos without video training, achieving SOTA on Video-MME via dense frame sampling.
-
Nomic Embed: Training a Reproducible Long Context Text Embedder
Nomic AI produced and open-sourced a reproducible 8192-context English text embedder that exceeds OpenAI Ada-002 and text-embedding-3-small performance on MTEB short-context and LoCo long-context benchmarks.
-
The Falcon Series of Open Language Models
Falcon-180B is a 180B-parameter open decoder-only model trained on 3.5 trillion tokens that approaches PaLM-2-Large performance at lower cost and is released with dataset extracts.
-
Llemma: An Open Language Model For Mathematics
Continued pretraining of Code Llama on Proof-Pile-2 yields Llemma, an open math-specialized LLM that beats known open base models on MATH and supports tool use plus formal proving out of the box.
-
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.
-
Efficient Streaming Language Models with Attention Sinks
StreamingLLM lets finite-window LLMs generalize to infinite-length sequences by retaining initial-token KV states as attention sinks, enabling stable streaming inference up to 4M tokens.
-
YaRN: Efficient Context Window Extension of Large Language Models
YaRN extends the context window of RoPE-based LLMs like LLaMA more efficiently than prior methods, using 10x fewer tokens and 2.5x fewer steps while surpassing state-of-the-art performance and enabling extrapolation beyond fine-tuning lengths.
-
ETC: Extreme Token Compression via Task-aware Visual Information Distillation in VLMs
ETC compresses visual tokens in VLMs to as few as one token via variational information distillation while retaining task performance on models like LLaVA-1.5-7B.
-
VIP-COP: Context Optimization for Tabular Foundation Models
VIP-COP is a black-box method that optimizes context for tabular foundation models by ranking and selecting high-value samples and features via online KernelSHAP regression, outperforming baselines on large high-dimensional data.
-
MemReread: Enhancing Agentic Long-Context Reasoning via Memory-Guided Rereading
MemReread improves agent long-context reasoning by triggering rereading on insufficient final memory to recover discarded indirect facts, outperforming baselines at linear complexity.
-
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.
-
Externalization in LLM Agents: A Unified Review of Memory, Skills, Protocols and Harness Engineering
LLM agent progress depends on externalizing cognitive functions into memory, skills, protocols, and harness engineering that coordinates them reliably.
-
Resolving the Robustness-Precision Trade-off in Financial RAG through Hybrid Document-Routed Retrieval
HDRR combines document-level semantic routing with scoped chunk retrieval to outperform both pure chunk-based retrieval and semantic file routing on the FinDER benchmark, delivering higher average scores, lower failure rates, and more perfect answers.
-
E2LLM: Encoder Elongated Large Language Models for Long-Context Understanding and Reasoning
E2LLM uses encoder-based soft prompt compression for long contexts to improve LLM reasoning on tasks like summarization and QA while maintaining efficiency.
-
DeepSeek-Coder: When the Large Language Model Meets Programming -- The Rise of Code Intelligence
DeepSeek-Coder open-source models trained on 2T code tokens with fill-in-the-blank pretraining achieve SOTA results among open models and surpass closed-source Codex and GPT-3.5 on code benchmarks.
-
MobileVLM : A Fast, Strong and Open Vision Language Assistant for Mobile Devices
MobileVLM achieves on-par performance with much larger vision-language models on standard benchmarks while delivering state-of-the-art inference speeds of 21.5 tokens per second on Snapdragon 888 CPU and 65.3 on Jetson Orin GPU.