AI Paper Atlas

Research map / Architectures and efficiency / Quantization

GSRQ: Gain-Shape Residual Quantization for Sub-1-bit KV Cache

Soosung Kim, Minjae Park, Eui-Young Chung, Jaeyong Chung

ICML 2026 · Architectures and efficiency › Quantization · approach: KV cache quantization · 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
replace standard K-means with GSKM (unweighted) (ablation)GSRQ residual quantization · WikiText-2 (0.375-bit)perplexity: 19.50 -> 12.77Improved
add raw gradient-norm weighting to GSKM (ablation)GSKM · WikiText-2 (0.375-bit)perplexity: 12.77 -> 13.57Hurt
add log-smoothed gradient weighting to GSKM (ablation)GSKM · WikiText-2 (0.375-bit)perplexity: 12.77 -> 12.49Improved
log-weighting instead of raw weighting vs. no weighting on keys (ablation)GSKM · WikiText-2 (0.375-bit), key perplexityperplexity: 6.87 -> 6.82Improved
GSRQ vs VQLLM at 1-bitLLaMA-3-8B-Instruct KV cache · LongBench averageaccuracy: 11.34 -> 33.54Improved
GSRQ vs VQLLM at 1-bitLLaMA-3-8B-Instruct KV cache · GSM8Kexact match accuracy: 2.27 -> 26.16Improved
GSRQ vs VQLLM at 2-bitMistral-7B-Instruct-v0.1 KV cache · GSM8Kexact match accuracy: 27.90 -> 28.13No clear effect
GSRQ vs VQLLM at 1-bitLLaMA-3-8B-Instruct KV cache · RULERaccuracy: 20.0 -> 72.0Improved
GSRQ vs VQLLM at 1-bitLLaMA-3-8B KV cache · C4 perplexityperplexity: 74.47 -> 8.92Improved
GSRQ vs VQLLM at 2-bit on Qwen3-8BQwen3-8B KV cache · Few-shot commonsense/math averageaccuracy: 36.44 -> 40.75Improved
GSRQ vs VQLLM at 1-bit on Qwen3-8BQwen3-8B KV cache · TriviaQA (LongBench)accuracy: 11.43 -> 8.17Hurt
GSRQ vs VQLLM at 2-bit on TruthfulQALLaMA-3-8B KV cache · TruthfulQAaccuracy: 51.05 -> 50.12Hurt

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