AI Paper Atlas

Research map / Large language models / Retrieval and knowledge

RetroLM: Retrieval-Augmented KVs for Long-Context Processing

Kun Luo, Zheng Liu, Shitao Xiao, Jiabei Chen, Hongjin Qian, Peitian Zhang, Shanshan Jiang, Bin Dong, Jun Zhao, Kang Liu

AAAI 2026 · Large language models › Retrieval and knowledge · approach: RAG robustness and evaluation · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add trained page retriever (Stage-1) (ablation)RetroLM · LongBench QA (6 tasks)average score: 29.6 -> 36.8Improved
add Stage-2 post-training on unsupervised text (ablation)RetroLM-Stage1 (Mistral) · LongBench (12 tasks)average score: 41.9 -> 43.5Improved
add Stage-2 post-training on unsupervised text (ablation)RetroLM-Stage1 (Llama-3) · LongBench (12 tasks)average score: 42.3 -> 44.4Improved
replace text-level RAG (BGE) with KV page retrievalMistral-7B · LongBench QA (6 tasks)average score: 31.4 -> 36.8Improved
replace heuristic SnapKV with trained page retrieverMistral-7B · LongBench (12 tasks)average score: 39.4 -> 41.9Improved
full-attention fine-tuning on same data instead of Stage-2 (ablation)Mistral-7B · LongBench QA (6 tasks)average score: 33.1 -> 33.9No clear effect
InfLLM on fine-tuned model instead of RetroLM (ablation)Mistral-7B · LongBench QA (6 tasks)average score: 32.5 vs 39.1Hurt
increase KV budget 512 -> 1024 -> 2048 (ablation)RetroLM-Stage1 · LongBench QA (6 tasks)average score: 32.3 -> 34.2 -> 36.8Improved
RetroLM-Stage2 vs SnapKV on ultra-long QAMistral-7B · InfiniteBench QAscore: 16.2 -> 20.2Improved
RetroLM-Stage2 vs SnapKV on ultra-long summarizationMistral-7B · InfiniteBench Summaryscore: 25.3 -> 29.2Improved
add Stage-2 trainingMistral-7B-Instruct · MMLUaccuracy: 62.5 -> 63.2No clear effect
add Stage-2 trainingMistral-7B-Instruct · GSM8Kaccuracy: 35.4 -> 34.6Hurt

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy