AI Paper Atlas

Research map / Architectures and efficiency / Quantization

AGoQ: Activation and Gradient Quantization for Memory-Efficient Distributed Training of LLMs

WenXiang Lin, Juntao Huang, LuHan Zhang, Laiyi Li, Xiang Bao, mengyang zhang, Bing Wang, Shaohuai Shi

ICML 2026 · Architectures and efficiency › Quantization · approach: Post-training quantization of LLMs · 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
use AGoQ instead of Megatron-LM on LLaMA2-13B at 80K sequenceLLaMA2-13B · seq 80K, 64 GPUsiteration time (ms): 149667 -> 111422Improved
use AGoQ instead of Megatron-LM and ZeRO-1 at 32KLLaMA2-13B · seq 32K, 64 GPUsiteration time (ms): 37635 -> 36568Improved
use AGoQ instead of COAT (24K)OLMo-1B · seq 24kmemory (MB): 94100 -> 66852Improved
use AGoQ instead of COAT (32K)OLMo-1B · seq 32kiteration time (ms): 8861 -> 8076Improved
add gradient quantization on top of activation-free optimizer quantization (O+G vs O) (ablation)AGoQ · LLaMA2-13B seq 12K GPUpeak memory (GB): 37.7 -> 35.3Improved
add activation quantization on top of O+G (ablation)AGoQ · LLaMA2-13B seq 12K GPUpeak memory (GB): 35.3 -> 22.3Improved
add activation quantization on top of O+G (NPU) (ablation)AGoQ · LLaMA2-13B seq 12K NPUpeak memory (GB): 40.5 -> 29.7Improved
add dynamic bit-width compensation (DBC) (ablation)AGoQ · sciq (LLaMA2-7B)accuracy (%): 71.20 -> 72.80Improved
add dynamic bit-width compensation (DBC) (ablation)AGoQ · arc_e (LLaMA2-7B)accuracy (%): 41.88 -> 41.58Hurt
add dynamic bit-width compensation (DBC) (ablation)AGoQ · arc_c (LLaMA2-7B)accuracy (%): 17.66 -> 18.34Improved
use layer-aware mixed-precision instead of uniform 4-bit activations (ablation)AGoQ · LLaMA2-7B pretrainingconvergenceImproved
fuse quantization/dequantization with GEMM (ablation)AGoQ · LLaMA3-8B GEMM shapesspeedup: 1.07x average; 1.05x end-to-endImproved

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