arXiv:2603. 26667v2 Announce Type: replace-cross Abstract: Retrieval-augmented generation (RAG) turns external documents into evidence for large language models.
By Xu Sun, Tongkai Xu, Baiheng Xie, Li Huang, Qiang Gao, Kunpeng Zhang
arXiv:2608.22137v1 Announce Type: new
Abstract: Modern language models and agents increasingly require persistent memory for complete codebases, long interaction histories, and heterogeneous enterpri...
By Xinyuan Song, Bowen Zhu, Hasibul Haque, Liang Zhao
arXiv:2607. 17486v1 Announce Type: cross Abstract: As large language models (LLMs) process increasingly longer prompts, computation and KV-cache memory costs have emerged as major bottlenecks in inference systems.
By Oteo Mamo, Hyunjin Yi, Joydhriti Choudhury, Shangqian Gao, Weikuan Yu
Retrieval-augmented generation (RAG) improves large language model (LLM) answer quality by grounding generation in external evidence, but processing retrieved contexts makes the prefill stage a dominant serving cost. RAG cache fusion reduces this cost by reusing precomputed key-value (KV) caches for retrieved chunks and selectively recomputing tokens under the current prompt.
ICICLE is an in‑context indexing framework that expands generative retrieval by supplying newly added documents as inference‑time evidence. It generates document identifiers using both parametric memory and context‑provided document‑docid pairs, employing a [COPY] routing mechanism, preference‑based calibration, and large‑context adaptation to separate context‑grounded retrieval from parametric retrieval. Experiments on MS MARCO and NQ320K demonstrate that ICICLE improves retrieval of new documents while retaining performance on previously indexed documents without retraining the model.
By Yu-Chen Den, Yung-Yu Shih, Zhi Rui Tam, Kuan-Yu Chen, Pu-Jen Cheng, Yun-Nung Chen, Eugene Yang
The paper introduces DEPT, a method that trains a single decoder-only large language model to both expand queries and encode documents for retrieval. By preserving document embeddings close to their initial cached values while allowing gradients to flow through the generator, DEPT stabilizes retrieval targets and enables efficient index reuse and online hard‑negative mining. Experiments on the BEIR benchmark with Qwen3‑4B‑Instruct‑2507 and LLaMA‑3.2‑3B‑Instruct show that DEPT outperforms training‑free, independently trained, and staged unified baselines, with ablations confirming the benefits of preservation, whitening, end‑to‑end expansion training, and online negatives.
By Jingyuan Wang, Richong Zhang, Zhijie Nie, Mingxin Li, Yanzhao Zhang