The upper-tail accumulation scale derived from the gap-counting function N_n sets the critical inverse temperature for softmax attention concentration, unifying prior conflicting laws as special cases of different N_n.
hub Mixed citations
YaRN: Efficient Context Window Extension of Large Language Models
Mixed citation behavior. Most common role is background (62%).
72
Pith papers citing it
Background
62% of classified citations
abstract
Rotary Position Embeddings (RoPE) have been shown to effectively encode positional information in transformer-based language models. However, these models fail to generalize past the sequence length they were trained on. We present YaRN (Yet another RoPE extensioN method), a compute-efficient method to extend the context window of such models, requiring 10x less tokens and 2.5x less training steps than previous methods. Using YaRN, we show that LLaMA models can effectively utilize and extrapolate to context lengths much longer than their original pre-training would allow, while also surpassing previous the state-of-the-art at context window extension. In addition, we demonstrate that YaRN exhibits the capability to extrapolate beyond the limited context of a fine-tuning dataset. Code is available at https://github.com/jquesnelle/yarn
hub tools
citation-role summary
background 11
method 4
other 1
citation-polarity summary
claims ledger
- abstract Rotary Position Embeddings (RoPE) have been shown to effectively encode positional information in transformer-based language models. However, these models fail to generalize past the sequence length they were trained on. We present YaRN (Yet another RoPE extensioN method), a compute-efficient method to extend the context window of such models, requiring 10x less tokens and 2.5x less training steps than previous methods. Using YaRN, we show that LLaMA models can effectively utilize and extrapolate to context lengths much longer than their original pre-training would allow, while also surpassing
co-cited works
representative citing papers
VINS-120K supplies the first large-scale set of instruction-image-edited-image triplets at ultra-high resolution together with an adaptation strategy that improves detail synthesis.
Hebatron is the first open-weight Hebrew MoE LLM adapted from Nemotron-3, reaching 73.8% on Hebrew reasoning benchmarks while activating only 3B parameters per pass and supporting 65k-token context.
Jordan-RoPE realizes a distance-modulated phase basis via non-semisimple Jordan blocks, generating features such as d e^{iωd} for relative positional encoding.
GVR uses previous-step Top-K predictions, pre-indexed stats, secant counting, and shared-memory verification to deliver 1.88x average speedup over radix-select while preserving bit-exact Top-K on DeepSeek-V3.2 workloads.
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.
TriAttention compresses KV cache by exploiting stable pre-RoPE Q/K concentration and trigonometric distance preferences to match full-attention reasoning accuracy with far lower memory and higher speed.
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.
FlashAttention-3 achieves 1.5-2x speedup on H100 GPUs for attention, reaching 740 TFLOPs/s (75% utilization) in FP16 and near 1.2 PFLOPs/s in FP8 while cutting numerical error by 2.6x versus baseline FP8 attention.
DeepSeek-V2 delivers top-tier open-source LLM performance using only 21B active parameters by compressing the KV cache 93.3% and cutting training costs 42.5% via MLA and DeepSeekMoE.
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.
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.
P2T distills reference patches into a latent process graph and uses it to select shortest effective trajectory segments from teacher rollouts, yielding up to 10.8 point Pass@1 gains on SWE-bench Verified with 15% lower inference cost using only 1.8k instances.
Repeating smaller datasets speeds up training via sampling biases that enable appropriate layer-wise growth, leading to compute savings over larger datasets across tasks and architectures.
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.
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.
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.
ZAYA1-8B is a reasoning MoE model with 700M active parameters that matches larger models on math and coding benchmarks and reaches 91.9% on AIME'25 via Markovian RSA test-time compute.
No model can achieve efficiency, compactness, and recall capacity scaling with sequence length at once, as any two imply a strict bound of O(poly(d)/log V) on recallable facts.
HyLo upcycles Transformer LLMs into hybrids with MLA and Mamba2/Gated DeltaNet blocks via staged training and distillation, extending context to 2M tokens and outperforming prior upcycled hybrids on long-context benchmarks.
TIGS detects backdoor-induced attention collapse in LLMs and applies content-aware tail-risk screening plus intrinsic geometric smoothing to suppress attacks while preserving normal performance.
citing papers explorer
-
A Unified Framework for Critical Scaling of Inverse Temperature in Self-Attention
The upper-tail accumulation scale derived from the gap-counting function N_n sets the critical inverse temperature for softmax attention concentration, unifying prior conflicting laws as special cases of different N_n.
-
VINS-120K: Ultra High-Resolution Image Editing with A Large-Scale Dataset
VINS-120K supplies the first large-scale set of instruction-image-edited-image triplets at ultra-high resolution together with an adaptation strategy that improves detail synthesis.
-
HEBATRON: A Hebrew-Specialized Open-Weight Mixture-of-Experts Language Model
Hebatron is the first open-weight Hebrew MoE LLM adapted from Nemotron-3, reaching 73.8% on Hebrew reasoning benchmarks while activating only 3B parameters per pass and supporting 65k-token context.
-
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.
-
Guess-Verify-Refine: Data-Aware Top-K for Sparse-Attention Decoding on Blackwell via Temporal Correlation
GVR uses previous-step Top-K predictions, pre-indexed stats, secant counting, and shared-memory verification to deliver 1.88x average speedup over radix-select while preserving bit-exact Top-K on DeepSeek-V3.2 workloads.
-
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.
-
TriAttention: Efficient Long Reasoning with Trigonometric KV Compression
TriAttention compresses KV cache by exploiting stable pre-RoPE Q/K concentration and trigonometric distance preferences to match full-attention reasoning accuracy with far lower memory and higher speed.
-
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.
-
FlashAttention-3: Fast and Accurate Attention with Asynchrony and Low-precision
FlashAttention-3 achieves 1.5-2x speedup on H100 GPUs for attention, reaching 740 TFLOPs/s (75% utilization) in FP16 and near 1.2 PFLOPs/s in FP8 while cutting numerical error by 2.6x versus baseline FP8 attention.
-
DeepSeek-V2: A Strong, Economical, and Efficient Mixture-of-Experts Language Model
DeepSeek-V2 delivers top-tier open-source LLM performance using only 21B active parameters by compressing the KV cache 93.3% and cutting training costs 42.5% via MLA and DeepSeekMoE.
-
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.
-
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.
-
From Patches to Trajectories: Privileged Process Supervision for Software-Engineering Agents
P2T distills reference patches into a latent process graph and uses it to select shortest effective trajectory segments from teacher rollouts, yielding up to 10.8 point Pass@1 gains on SWE-bench Verified with 15% lower inference cost using only 1.8k instances.
-
Less Data, Faster Training: repeating smaller datasets speeds up learning via sampling biases
Repeating smaller datasets speeds up training via sampling biases that enable appropriate layer-wise growth, leading to compute savings over larger datasets across tasks and architectures.
-
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.
-
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.
-
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.
-
ZAYA1-8B Technical Report
ZAYA1-8B is a reasoning MoE model with 700M active parameters that matches larger models on math and coding benchmarks and reaches 91.9% on AIME'25 via Markovian RSA test-time compute.
-
The Impossibility Triangle of Long-Context Modeling
No model can achieve efficiency, compactness, and recall capacity scaling with sequence length at once, as any two imply a strict bound of O(poly(d)/log V) on recallable facts.
-
Long-Context Aware Upcycling: A New Frontier for Hybrid LLM Scaling
HyLo upcycles Transformer LLMs into hybrids with MLA and Mamba2/Gated DeltaNet blocks via staged training and distillation, extending context to 2M tokens and outperforming prior upcycled hybrids on long-context benchmarks.
-
Defusing the Trigger: Plug-and-Play Defense for Backdoored LLMs via Tail-Risk Intrinsic Geometric Smoothing
TIGS detects backdoor-induced attention collapse in LLMs and applies content-aware tail-risk screening plus intrinsic geometric smoothing to suppress attacks while preserving normal performance.
-
Reducing Detail Hallucinations in Long-Context Regulatory Understanding via Targeted Preference Optimization
DetailDPO cuts detail-level hallucination errors in LLMs on long regulatory documents by 42-61% using targeted contrastive pairs on a new 13,000-pair benchmark.
-
Democratizing Tool Learning with Environments Fully Simulated by a Free 8B Language Model
TRUSTEE uses an 8B LM to simulate complete dynamic environments for RL-based tool learning and outperforms baselines that require extra external resources.
-
OPSDL: On-Policy Self-Distillation for Long-Context Language Models
OPSDL improves long-context LLM performance by having the model self-distill from its short-context capability using point-wise reverse KL divergence on generated tokens, outperforming SFT and DPO on benchmarks without harming short-context abilities.
-
Sensitivity-Positional Co-Localization in GQA Transformers
In Llama 3.1 8B, task-sensitive layers cluster late while RoPE adaptation is strongest early, yet applying both adaptations only to sensitivity-identified layers outperforms other layer choices by 4-16 points on MMLU, GPQA, HumanEval+, MATH, MGSM and ARC.
-
In-Place Test-Time Training
In-Place TTT adapts LLM MLP projection matrices at test time with a next-token-aligned objective and chunk-wise updates, enabling better long-context performance as a drop-in enhancement.
-
From SWE-ZERO to SWE-HERO: Execution-free to Execution-based Fine-tuning for Software Engineering Agents
A two-stage SFT pipeline distills execution-free then execution-based trajectories from a 480B model into smaller Qwen2.5-Coder agents, yielding 62.2% resolution on SWE-bench Verified and 44.1% zero-shot on the multilingual version.
-
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.
-
WebExpert: domain-aware web agents with critic-guided expert experience for high-precision search
WebExpert improves exact-match accuracy by 1.5-3.6 points on GAIA, GPQA, HLE, and WebWalkerQA benchmarks via experience retrieval, automatic facet induction, and preference-optimized planning.
-
Olmo 3
Olmo 3 delivers fully open 7B and 32B language models with complete training artifacts, positioning the 32B variant as the strongest open thinking model released to date.
-
Kimi Linear: An Expressive, Efficient Attention Architecture
Kimi Linear hybridizes linear attention with a new KDA module to beat full attention on tasks while slashing KV cache by 75% and speeding decoding up to 6x.
-
StreamingVLM: Real-Time Understanding for Infinite Video Streams
StreamingVLM enables stable real-time understanding of infinite video streams at up to 8 FPS using a streaming KV cache and aligned SFT on overlapped chunks, with a 66.18% win rate over GPT-4O mini on a new two-hour video benchmark.
-
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.
-
Short window attention enables long-term memorization
Short sliding windows in hybrid attention-xLSTM models boost long-context performance by encouraging long-term memory use, and stochastic window sizing improves both short and long tasks.
-
EditVerse: Unifying Image and Video Editing and Generation with In-Context Learning
EditVerse unifies image and video editing and generation in one transformer model via unified token sequences and in-context learning, trained jointly on curated video editing data plus image/video corpora and evaluated on a new instruction-based benchmark.
-
Accelerating Prefilling via Decoding-time Contribution Sparsity
TriangleMix exploits decoding-time contribution sparsity via a training-free static attention pattern to accelerate LLM prefilling with nearly lossless performance.
-
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.
-
Does Math Reasoning Improve General LLM Capabilities? Understanding Transferability of LLM Reasoning
Math reasoning gains in LLMs rarely transfer to general domains; RL tuning generalizes while SFT causes forgetting and representation drift.
-
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.
-
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.
-
Muon is Scalable for LLM Training
Muon optimizer with weight decay and update scaling achieves ~2x efficiency over AdamW for large LLMs, shown via the Moonlight 3B/16B MoE model trained on 5.7T tokens.
-
Qwen2.5-1M Technical Report
Qwen2.5-1M models reach 1M token context with improved long-context performance, no short-context loss, and 3-7x prefill speedup via open inference optimizations.
-
DataComp-LM: In search of the next generation of training sets for language models
DCLM-Baseline dataset lets a 7B model reach 64% 5-shot MMLU accuracy after 2.6T tokens, beating prior open-data models by 6.6 points on MMLU with 40% less compute.
-
DeepSeek-Coder-V2: Breaking the Barrier of Closed-Source Models in Code Intelligence
An open-source MoE code model matches GPT-4 Turbo on coding and math benchmarks while expanding to 338 languages and 128K context length.
-
MiniCPM: Unveiling the Potential of Small Language Models with Scalable Training Strategies
MiniCPM 1.2B and 2.4B models reach parity with 7B-13B LLMs via model wind-tunnel scaling and a WSD scheduler that yields a higher optimal data-to-model ratio than Chinchilla scaling.
-
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.
-
DynaFlow: Transparent and Flexible Intra-Device Parallelism via Programmable Operator Scheduling
DynaFlow enables transparent intra-device parallelism in ML systems by separating model definition from execution scheduling, integrating into 6 frameworks with up to 1.29x throughput gains and minimal code changes.