AI Paper Atlas

Research map / Architectures and efficiency / Quantization

CoCoQuant: Breaking the Bandwidth Wall via Co-Optimized Communication and Computation Quantization

Haojie Duanmu, Jifeng Ding, Size Zheng, Xuegui Zheng, Jiangfei Duan, Xingcheng Zhang, Li-Wen Chang, Xin Liu, Dahua Lin

ICML 2026 · Architectures and efficiency › Quantization · approach: Post-training quantization of LLMs

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
quantize only computation (W4A4C16)Llama3.3-70B-Instruct · 8x RTX 4090speedup: 1.31x speedup with PPL 3.63 -> 7.30Hurt
add communication quantization with mixed precision (CoCoQuant 2) instead of W8A8C8Llama3.3-70B-Instruct · 8x RTX 4090speedup: 1.82x -> 2.09xImproved
use CoCoQuant 1 instead of uniform W4A4C4*Llama3-8B-Instruct · WikiText-2perplexity: 9.21 -> 7.87Improved
use CoCoQuant 1 instead of uniform W4A4C4*Qwen3-MoE-235B · 8x RTX 4090speedup: 2.01x -> 2.92xImproved
apply W4A4C16 to MoELlama4-Scout-109B · WikiText-2perplexity: 7.76 -> 457.96Hurt
add precision-aligned graph rewriting (PGR) (ablation)CoCoQuant W4A4C4 · Llama3.3-70B WikiText-2perplexity: 6.29 -> 4.87Improved
add precision-aligned graph rewriting (PGR) (ablation)CoCoQuant W4A4C4 · Llama3.3-70Bspeedup: 2.78x -> 3.52xImproved
add precision-aligned graph rewriting (PGR) (ablation)CoCoQuant W8A8C8 · Llama3.3-70B WikiText-2perplexity: 3.76 -> 3.66Improved
add precision-aligned graph rewriting (PGR) (ablation)CoCoQuant W8A8C8 · Llama3.3-70Bspeedup: 1.74x -> 2.05xImproved
use ILP mixed precision instead of uniform at W4A4C4 latency (ablation)CoCoQuant · GSM8K Llama3-8B-Instructaccuracy (%): 46.63 -> 73.62Improved
use ILP mixed precision instead of uniform at W8A8C8 latency (ablation)CoCoQuant · GSM8K Llama3-8B-Instructaccuracy (%): 76.12 -> 76.22No clear effect
use ILP mixed precision instead of uniform at W4A4C4 latency (ablation)CoCoQuant · GSM8K Llama3-8Baccuracy (%): 21.23 -> 43.82Improved

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