AI Paper Atlas

Research map / Optimization / Distributed and compressed

LoCo: Low-Bit Communication Adaptor for Large-Scale Model Training

Xingyu Xie, Zhijie Lin, Kim-Chuan Toh, Pan Zhou

TPAMI 2025 · Optimization › Distributed and compressed · approach: Gradient quantization and sparsification · 13 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
add error feedback (ablation)LoCo · Alpaca-GPT4 fine-tuneaverage accuracy: 67.4% -> 67.5%No clear effect
add moving average on compression error (ablation)LoCo · Alpaca-GPT4 fine-tuneaverage accuracy: 67.5% -> 67.7%Improved
add 8-bit compression of stored error (ablation)LoCo · Alpaca-GPT4 fine-tuneaverage accuracy: 67.8% -> 67.9%No clear effect
add periodic error reset (every 512 iterations) (ablation)LoCo · Alpaca-GPT4 fine-tuneaverage accuracy: 67.7% -> 67.9%Improved
reset error every 128 instead of 512 iterations (ablation)LoCo · Alpaca-GPT4 fine-tuneaverage accuracy: 67.9% -> 68.1%Improved
replace Zero++ 4-bit with LoCo 4-bitZero++ · LLaMA2-7B Alpaca-GPT4average accuracy: 44.3% -> 45.6%Improved
replace 0/1 Adam 4-bit with LoCo 4-bit0/1 Adam · LLaMA2-7B Alpaca-GPT4average accuracy: 44.8% -> 45.6%Improved
use 4-bit LoCo communication instead of 16-bit AdamAdam · LLaMA2-7B RedPajama fine-tuneaverage accuracy: 39.75% -> 39.70%No clear effect
use 4-bit LoCo instead of PowerSGDPowerSGD · LLaMA2-7B LoRAaverage accuracy: 45.4% -> 45.7%Improved
use 4-bit LoCo communication in from-scratch pre-trainingAdam · Sky-MoE 300B tokenstraining loss: 2.105 -> 2.102No clear effect
combine LoCo with Zero++ (LoCo-Zero++)Zero++ · LLaMA2-0.8B and 8B pre-trainingtraining lossImproved
increase GPU count from 32 to 128LoCo · LLaMA2-13B Megatron-LM A800speedup (%): 24.58% -> 42.24%Improved

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