Mammoth: benchmark continual-learning methods

Mammoth

Mammoth is a modular, open-source Python framework built on PyTorch for developing and benchmarking continual-learning research methods. Its repository describes more than 70 methods and 20 datasets, and provides extension points for models, datasets, and training strategies.

Independent OpenViST project profile, based on linked sources.

Understand the project

Practical overview

When it fits

Use Mammoth when you want a code-based research workflow for running and extending continual-learning experiments, rather than a hosted model service.

Who it is for

Researchers comparing continual-learning methods across datasets.
Python developers adding a model, dataset, or training strategy to a continual-learning research workflow.

Documented uses

Run a continual-learning method on a dataset with specified hyperparameters.
Use a model’s saved best-configuration settings for an experiment.
Extend the framework with custom models or datasets.

Before you choose

Requirements & limits

The README describes installation from a cloned repository and says Mammoth is not available on PyPI.

The framework is built on PyTorch and written in Python.
The README specifies PyTorch 2.1.0 or later for scaled_dot_product_attention and describes a code adjustment for environments that cannot support it.
The repository metadata specifies an MIT license.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Run a documented DER++ benchmark

Run DER++ on the sequential CIFAR-100 dataset using the README’s example hyperparameters.

Before you start

A cloned Mammoth repository and Python environment with the repository requirements installed.
PyTorch 2.1.0 or later for scaled_dot_product_attention, unless using the documented code adjustment.

Steps

Install dependencies with the command below.
Documented command
```
pip install -r requirements.txt
```

From the repository, run the command below.

Documented command

python main.py --model derpp --dataset seq-cifar100 --alpha 0.5 --beta 0.5 --lr 0.001 --buffer_size 500

Expected outcome

The command starts a DER++ experiment on seq-cifar100 with the specified buffer size and hyperparameters.

Guide limitations

The README documents the command, but does not establish a runtime result or expected metric.
The stated PyTorch requirement applies to scaled_dot_product_attention; the README describes editing backbone/vit.py if that requirement cannot be met.

Documentation 1 ↗

Evidence and transparency

Sources & review

Official repository ↗

Source record details

Source availability: Active
Sources checked: 6 November 2026
Original website title: aimagelab/mammoth · Official homepage title · retained in its source language

How we review sources Suggest a correction