54 articles tagged with #long-context. AI-curated summaries with sentiment analysis and key takeaways from 50+ sources.
Researchers introduce CSAttention, a training-free sparse attention method that accelerates LLM inference by 4.6x for long-context applications. The technique optimizes the offline-prefill/online-decode workflow by precomputing query-centric lookup tables, enabling faster token generation without sacrificing accuracy even at 95% sparsity levels.
Researchers propose SWAA (Sliding Window Attention Adaptation), a toolkit that enables efficient long-context processing in large language models by adapting full attention models to sliding window attention without expensive retraining. The solution achieves 30-100% speedups for long context inference while maintaining acceptable performance quality through four core strategies that address training-inference mismatches.
Researchers propose ARKV, a new framework for managing memory in large language models that reduces KV cache memory usage by 4x while preserving 97% of baseline accuracy. The adaptive system dynamically allocates precision levels to cached tokens based on attention patterns, enabling more efficient long-context inference without requiring model retraining.
Researchers introduce FlashPrefill, a new framework that dramatically improves Large Language Model efficiency during the prefilling phase through advanced sparse attention mechanisms. The system achieves up to 27.78x speedup on long 256K sequences while maintaining 1.71x speedup even on shorter 4K contexts.
Researchers introduce ASEntmax, a new attention mechanism for transformer models that uses sparse attention with learnable temperature parameters. This approach significantly outperforms traditional softmax attention, achieving up to 1000x length extrapolation on synthetic tasks and better long-context performance in language modeling.
Researchers introduce RMAAT (Recurrent Memory Augmented Astromorphic Transformer), a new architecture inspired by brain astrocyte cells that addresses the quadratic complexity problem in Transformer models for long sequences. The system uses recurrent memory tokens and adaptive compression to achieve linear complexity while maintaining competitive accuracy on benchmark tests.
MiniCPM-SALA introduces a 9B-parameter hybrid language model architecture that combines sparse and linear attention mechanisms to handle ultra-long contexts up to 1M tokens. The model achieves 3.5x faster inference than full-attention models while reducing training costs by 75% through a continual training framework that transforms existing Transformer models.
Researchers introduce S2O, a new sparse attention method that uses online permutation and early stopping to dramatically improve AI model efficiency. The technique achieves 3.81x end-to-end speedup on Llama-3.1-8B with 128K context while maintaining accuracy.
OpenAI has announced GPT-5.2, their most advanced frontier AI model designed for professional applications. The model features enhanced reasoning capabilities, long-context understanding, coding abilities, and vision functionality, available through ChatGPT and OpenAI API for improved agentic workflows.
Researchers introduce Membox, a hierarchical memory architecture for LLM agents that organizes dialogue history by topic continuity rather than semantic proximity. The system uses Topic Loom to group related turns and Trace Weaver to link events across sessions, achieving 13-19 percentage point F1 improvements over existing memory systems like Mem0 and A-MEM.
Researchers compare retrieval-augmented generation (RAG) versus long-context prompting for document-grounded AI applications, finding that while long-context achieves higher accuracy (73.1% vs 65.4%), it incurs a 26x higher token cost. The study frames this trade-off as an 'epistemic accuracy' versus computational expense frontier, with significant implications for resource-constrained organizations.
Researchers introduce Blurry Window Attention (BLA), a novel attention mechanism that addresses the quadratic complexity and memory limitations of traditional Transformer models by reconstructing sparse key-value history through Dirichlet kernel interpolation. BLA demonstrates 8x state efficiency improvements over sliding window attention while maintaining competitive performance on information retrieval tasks, positioning it as a viable alternative for long-context language modeling.
Researchers introduce DySink, a novel framework for autoregressive long video generation that dynamically selects relevant historical frames instead of using static early-frame anchors. The method addresses the problem of outdated context degrading video quality and introduces a sink anomaly gate to prevent content collapse, demonstrating improvements in temporal consistency for minute-long videos.
Researchers introduce SemanticSeg, a large semantic segmentation dataset, and block distillation framework to improve block attention mechanisms for long-context language models. The approach uses a frozen full-attention teacher to train block-attention students more efficiently, addressing key challenges in KV cache reuse for applications like RAG.
Researchers propose SISA (SSM-Informed Softmax Attention), a hybrid architecture that integrates state space model importance signals directly into transformer attention mechanisms at the score level. The approach achieves superior performance on language modeling benchmarks, particularly excelling at long-context retrieval tasks while maintaining computational efficiency through standard operations.
Researchers introduce RefMem-Bench, a new benchmark for evaluating reflective memory in AI dialogue systems, along with REMIND, a framework designed to improve how models synthesize fragmented information across long conversations. The work addresses a gap in existing benchmarks that measure only explicit recall rather than higher-level reasoning and interpretation.
Researchers propose a unified framework for long-form egocentric video understanding that separates reasoning into semantic and visual evidence streams, achieving competitive results on the HD-EPIC-VQA benchmark. The approach addresses fundamental limitations in how multimodal language models process extended video content by combining procedural structure extraction with fine-grained object grounding.
Researchers introduce MGRetrieval, a novel retrieval strategy for long-term dialogue agents that uses semantic memory structures to guide multi-step retrieval rather than one-shot approaches. The method improves performance on dialogue benchmarks by 8-11% while maintaining computational efficiency, addressing a key limitation in LLM-based conversational systems.
Researchers present a diagnostic framework for evaluating KV cache eviction selectors in large language models, identifying three failure modes and demonstrating that value-aware ranking combined with evidence recovery achieves 72.6% accuracy on positive-margin test cases. The work addresses a critical bottleneck in long-context LLM inference by revealing why compression strategies succeed or fail.
Researchers present a theoretical analysis of how transformer attention mechanisms scale with context length, identifying a critical threshold where attention shifts from uniform averaging to focusing on individual keys. The findings establish that this transition point depends on local geometric properties of the key distribution rather than global features, with implications for understanding transformer behavior at extreme context lengths.
Researchers introduce RecaLLM, a post-trained language model that addresses the 'lost-in-thought' phenomenon where retrieval performance degrades during extended reasoning chains. The model interleaves explicit in-context retrieval with reasoning steps and achieves strong performance on long-context benchmarks using training data significantly shorter than existing approaches.
Researchers have developed LookaheadKV, a new framework that significantly improves memory efficiency in large language models by intelligently evicting less important cached data. The method achieves superior accuracy while reducing computational costs by up to 14.5x compared to existing approaches, making long-context AI tasks more practical.
Researchers developed a new token reduction method for hybrid vision-language models that process long videos, achieving 3.8-4.2x speedup while retaining only 25% of visual tokens. The approach uses progressive reduction and unified scoring for both attention and Mamba blocks, maintaining near-baseline accuracy on long-context video benchmarks.
Researchers introduced SciTrek, a new benchmark for testing large language models' ability to perform numerical reasoning across long scientific documents. The benchmark reveals significant challenges for current LLMs, with the best model achieving only 46.5% accuracy at 128K tokens, and performance declining as context length increases.
Researchers introduce ReMemR1, a new approach to improve large language models' ability to handle long-context question answering by integrating memory retrieval into the memory update process. The system enables non-linear reasoning through selective callback of historical memories and uses multi-level reward design to strengthen training.
