AI Paper Atlas

Research map / Optimization / Distributed and compressed

1-bit Adam: Communication Efficient Large-Scale Training with Adam’s Convergence Speed

Hanlin Tang, Shaoduo Gan, Ammar Ahmad Awan, Samyam Rajbhandari, Conglong Li, Xiangru Lian, Ji Liu, Ce Zhang, Yuxiong He

ICML 2021 · Optimization › Distributed and compressed · approach: Gradient quantization and sparsification · 14 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 APMSqueeze instead of uncompressed BertAdamBERT-Base and BERT-Large fine-tuning · GLUE development setsscore: RTE 68.2 -> 69.0; MRPC 84.8 -> 84.8; CoLA 56.8 -> 55.6; SST-2 91.8 -> 91.6; QNLI 90.9 -> 90.8; QQP 90.9 -> 90.9; MNLI-(mMixed
Use APMSqueeze instead of uncompressed BertAdamBERT-Large fine-tuning · GLUE development setsscore: RTE 70.3 -> 70.4; MRPC 86.0 -> 86.1; CoLA 60.3 -> 62.0; SST-2 93.1 -> 93.8; QNLI 92.2 -> 91.9; QQP 91.4 -> 91.5; MNLI-(mMixed
Use APMSqueeze rather than AdamBERT-Large pre-training · BERT-Large sequence length 128, 64 GPUstraining time: 174.3 hours -> 51.5 hoursImproved
Compress momentum with error compensation after Adam warm-up and freeze the variance termBERT pre-training · BERT-Base and BERT-Large, both sequence length phasescommunication volume: reduced into 6% of the originalImproved
Use APMSqueeze rather than AdamBERT-Large pre-training · BERT-Large sequence length 128throughput: 3.3xImproved
Use APMSqueeze rather than AdamSQuAD fine-tuning · SQuAD 1.1throughput: up to 2.9xImproved
Use APMSqueeze instead of AdamBERT fine-tuning · SQuAD 1.1F1: 93.33 -> 93.32No clear effect
Compress gradients directly and keep updating the variance term, rather than compressing momentum and freezing variance (ablation)Adam(1-bit Naive) · CIFAR-10 with ResNet-18convergence speedHurt
Use APMSqueeze with 32-bit uncompressed momentum (ablation)APMSqueeze · CIFAR-10 with ResNet-18convergence speedNo clear effect
Replace APMSqueeze with SGDResNet-18 · CIFAR-10convergence speedHurt
Use APMSqueeze rather than uncompressed AdamDCGAN · CelebAtraining accuracyNo clear effect
Use APMSqueeze rather than BertAdamBERT pre-training · BERT-Base and BERT-Large, sequence length 128 and 512sample-wise convergence speedNo clear effect

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