A comprehensive index of artificial intelligence and machine-learning research with AI-generated summaries, citation metrics, and direct links to papers and code.
Alex L. Zhang, T. Kraska, Omar Khattab
Recursive Language Models (RLMs) enable LLMs to process prompts far beyond their context windows by recursively examining and calling themselves over prompt snippets, outperforming existing long-context methods.
Gyuwan Kim, Cheoneum Park, Tao Yang
CoinRAG optimizes the Pareto frontier for long-context RAG by reusing fine-grained, query-relevant KV caches to reduce latency and improve answer quality.
Unknown
ChunkKV is a novel KV cache compression method that retains the most informative semantic chunks to improve efficiency in long-context LLM inference.
Unknown
Longred introduces restoration distillation to mitigate short-text degradation in long-context LLMs by restoring original short-text performance during long-context adaptation.
Yida Lu, Jiale Cheng, Zhexin Zhang, et al.
This paper introduces LongSafety, a benchmark and framework for evaluating the safety of large language models in long-context scenarios, revealing that safety risks increase with context length.
Unknown
This paper proposes a focused learning approach to reduce distraction in long-context language models, improving their ability to utilize relevant information.
Unknown
Longproc introduces a benchmark for evaluating long-context language models on procedural generation tasks, addressing the lack of long-range dependency testing in existing benchmarks.
Unknown
This paper explores how long-context language models can perform in-context retrieval and reasoning, potentially simplifying RAG pipelines.
Unknown
This paper introduces Longrecipe, a recipe for efficient long-context generalization in large language models, addressing limited effective context window during pretraining.
Indraneil Paul, Falko Helm, Goran Glavaš, et al.
OctoLong introduces a context engineering pipeline that curates dependency-rich code contexts to enhance long-context modeling in language models.
Unknown
This paper explores how task type influences KV cache compression and proposes DynamicKV, a task-aware adaptive compression method for long-context LLMs.
Unknown
This paper explores in-context learning at extreme scale with long-context models, studying properties of ICL and long-context behavior across multiple datasets.