AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

RaBitQCache: Rotated Binary Quantization for KVCache in Long Context LLM Inference

Wenhao Li, Jinhao Dong, Hailin Zhang, Wenhang Shi, WEI LU, Xiaoyong Du

ICML 2026 · Architectures and efficiency › Attention and transformers · approach: KV cache eviction and compression · 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 centroid re-centering before quantization (ablation)RaBitQCache · LongBench LLaMA-3.1-8Bgeneration score: 50.25 -> 50.63Improved
use adaptive top-p (p=0.95) vs full attentionRaBitQCache · LongBench LongChatgeneration score: 35.78 -> 36.21No clear effect
use adaptive top-p (p=0.95) vs full attentionRaBitQCache · LongBench LLaMA-3.1-8Bgeneration score: 50.58 -> 50.6No clear effect
use RaBitQCache instead of Quest-1024LongBench LLaMA-3.1-8B · LongBench LLaMA-3.1-8Bgeneration score: 46.52 -> 50.6Improved
use RaBitQCache instead of Double Sparse-256LongChat · LongBench LongChatgeneration score: 3.75 -> 36.21Improved
use RaBitQCache vs Quest-4096RaBitQCache · LongBench LongChatgeneration score: 36.21 -> 36.28Hurt
use RaBitQCache instead of Quest on 70BRaBitQCache · LongBench LLaMA-3.1-70Bgeneration score: 50.69 -> 54.58Improved
use RaBitQCache instead of Quest on GSM8KRaBitQCache · GSM8Kaccuracy: 0.70 -> 0.77Improved
use RaBitQCache instead of PyramidKV/SnapKVRaBitQCache · LongBench LLaMA-3.1-8Bgeneration score: 45.09 -> 50.63Improved
use optimized warp kernel vs naive Int4DotBinary (ablation)Int4DotBinary kernel · operator microbenchmarkspeedup: up to 3.43xImproved
use RaBitQCache decoding vs full attentionRaBitQCache · decoding latency 10k-32kspeedup: up to 3.88xImproved
use RaBitQCache end to end vs full precisionRaBitQCache · end-to-end latencyspeedup: 2.16xImproved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 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