AI Paper Atlas

Research map / Architectures and efficiency / Quantization

Channel-Aware Mixed-Precision Quantization for Efficient Long-Context Inference

Chengxi Liao, Zeyi Wen

ICLR 2026 · Architectures and efficiency › Quantization · approach: KV cache quantization

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 retrieval channel 4-bit allocation to 2-bit baseline (ablation)ChanMix2 · RULER (average) Llama-3.1accuracy: 75.40 -> 81.50Improved
add retrieval channel allocation to 2-bit baseline (ablation)ChanMix2 · RULER (average) Mistralaccuracy: 72.13 -> 86.12Improved
add outlier channel 3-bit compensation to 2-bit baseline (ablation)ChanMix2 · RULER (average) Llama-3.1accuracy: 75.40 -> 81.13Improved
add outlier compensation to 2-bit baseline (ablation)ChanMix2 · RULER (average) Mistralaccuracy: 72.13 -> 84.17Improved
add outlier channels on top of retrieval channels (ablation)ChanMix2+R · RULER (average) Llama-3.1accuracy: 81.50 -> 82.16Improved
add outlier channels on top of retrieval channels (ablation)ChanMix2+R · RULER (average) Mistralaccuracy: 86.12 -> 86.35Improved
bit reallocation versus uniform 2-bit at group 128, residual 32 (ablation)ChanMix · RULER Llama-3.1accuracy: 69.78 -> 78.02Improved
larger group size 128 instead of 32 at residual 32 (ablation)ChanMix · RULER Llama-3.1accuracy: 83.92 -> 78.02Hurt
vary residual length 32/64/128 at group 32 (ablation)ChanMix · RULER Llama-3.1accuracy: 83.92, 84, 83.87No clear effect
channel-aware mixed precision versus KIVIChanMix · RULER Llama-3.1accuracy: 76.51 -> 82.16Improved
ChanMix versus KIVI on short-context GSM8KChanMix · GSM8K Llama-3.1accuracy: 74.75 -> 73.54Hurt
ChanMix versus KIVI on reasoning AIME24ChanMix · AIME24 DS-R1-Llamapass@1: 26.67 -> 26.67No clear effect

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