AI Paper Atlas

Research map / Architectures and efficiency / Quantization

PM-KVQ: Progressive Mixed-precision KV Cache Quantization for Long-CoT LLMs

Tengxuan Liu, Shiyao Li, Jiayi Yang, Tianchen Zhao, Feng Zhou, Xiaohui Song, Guohao Dai, Shengen Yan, Huazhong Yang, Yu Wang

ICLR 2026 · Architectures and efficiency › Quantization · approach: KV cache quantization · 19 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
use equivalent right shift instead of direct right shift when lowering bit-width (ablation)PM-KVQ · AIME-2024 (DeepSeek-LLaMA-8B, 4-bit)pass@1: 12.08 -> 38.33Improved
use equivalent right shift instead of modified right shift (ablation)PM-KVQ · AIME-2024 (DeepSeek-LLaMA-8B, 4-bit)pass@1: 28.75 -> 38.33Improved
add positional interpolation (s=4) to 2,048-token calibration (ablation)PM-KVQ · AIME-2024-I (DeepSeek-LLaMA-8B)pass@1: 46.67 -> 48.33Improved
increase position scaling factor from 4 to 16 (ablation)PM-KVQ · AIME-2024-I (DeepSeek-LLaMA-8B)pass@1: 48.33 -> 46.67Hurt
add progressive quantization with uniform Fbit versus KIVI 2-bitPM-KVQ · AIME-2024 (DeepSeek-Qwen-7B)pass@1: 32.08 -> 40.00Improved
add block-wise memory allocation (BS=32 mixed 2/4) over uniform 2-bit (ablation)PM-KVQ · AIME-2024 (DeepSeek-Qwen-7B)pass@1: 40.00 -> 40.21No clear effect
add block-wise memory allocation (BS=12) over uniform 2-bit (BS=16) (ablation)PM-KVQ · AIME-2024 (DeepSeek-Qwen-14B)pass@1: 63.33 -> 67.71Improved
add block-wise memory allocation (BS=12) over uniform 2-bit (BS=16) (ablation)PM-KVQ · AIME-2024 (QwQ-32B)pass@1: 67.29 -> 66.46Hurt
replace KIVI 2-bit with PM-KVQ on 70B modelDeepSeek-R1-Distill-LLaMA-70B · AIME-2024pass@1: 51.88 -> 64.79Improved
replace KIVI with PM-KVQ on 14B modelDeepSeek-R1-Distill-Qwen-14B · AIME-2024pass@1: 48.13 -> 67.71Improved
replace KVTuner with PM-KVQDeepSeek-R1-Distill-Qwen-7B · AIME-2024pass@1: 34.17 -> 40.21Improved
replace KIVI with PM-KVQPM-KVQ vs KIVI throughput · Throughput 32K, Qwen-7Btokens/s: 284.10 -> 269.51Hurt

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