# Stop Wasting My Time! Saving Days of ImageNet and BERT Training with Latest Weight Averaging

Source: [Stop Wasting My Time! Saving Days of ImageNet and BERT Training with Latest Weight Averaging](<https://www.sota2.com/research/paper/68fa6d8c1e209131cea3a866>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2025-10-26T00:37:22.464Z

Authors: Jean Kaddour

Published: 2022-09-29T17:46:13.000Z

## Abstract

Training vision or language models on large datasets can take days, if not weeks. We show that averaging the weights of the k latest checkpoints, each collected at the end of an epoch, can speed up the training progression in terms of loss and accuracy by dozens of epochs, corresponding to time savings up to ~68 and ~30 GPU hours when training a ResNet50 on ImageNet and RoBERTa-Base model on WikiText-103, respectively. We also provide the code and model checkpoint trajectory to reproduce the results and facilitate research on reusing historical weights for faster convergence.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2209.14981>)

[PDF](<https://arxiv.org/pdf/2209.14981.pdf>)