#3232nd of 68 in Training and fine-tuning

lightly

PyTorch library for self-supervised vision pretraining and latent world models

Stars
3.8k
License
MIT
Last commit
Nov 2026
Last release
Nov 2026
Language
Python
Stars, last 29 weeks+121
25 Apr 2026: 3,710 stars2 May 2026: 3,715 stars9 May 2026: 3,717 stars16 May 2026: 3,723 stars23 May 2026: 3,730 stars30 May 2026: 3,739 stars6 Jun 2026: 3,746 stars13 Jun 2026: 3,750 stars20 Jun 2026: 3,751 stars27 Jun 2026: 3,756 stars4 Jul 2026: 3,761 stars11 Jul 2026: 3,767 stars18 Jul 2026: 3,770 stars25 Jul 2026: 3,775 stars1 Aug 2026: 3,779 stars8 Aug 2026: 3,786 stars15 Aug 2026: 3,788 stars22 Aug 2026: 3,790 stars29 Aug 2026: 3,795 stars5 Sept 2026: 3,798 stars12 Sept 2026: 3,803 stars19 Sept 2026: 3,809 stars26 Sept 2026: 3,813 stars3 Oct 2026: 3,815 stars10 Oct 2026: 3,817 stars17 Oct 2026: 3,820 stars24 Oct 2026: 3,821 stars31 Oct 2026: 3,825 stars7 Nov 2026: 3,830 stars14 Nov 2026: 3,831 stars

3,710 on 25 Apr 20263,831 on 14 Nov 2026

From GitHub's daily star counts.

Checked on GitHub.

Overview

LightlySSL is a Python framework of modular building blocks (loss functions, model heads, transforms) for self-supervised pretraining of computer vision models. It ships PyTorch and PyTorch Lightning examples for about 25 methods, including SimCLR, MoCo, BYOL, DINO, DINOv2, MAE and LeJEPA, plus LeWM for training world models and planning with them. Custom backbones are supported, and distributed training goes through PyTorch Lightning.

Who it is for: ML engineers pretraining vision backbones with self-supervised methods

Strengths

  • Covers MoCo (2019) through LeJEPA and Pixio (2025), each with docs and Colab notebooks
  • Modular losses and heads that can be mixed with custom backbones
  • PyTorch and PyTorch Lightning examples, including distributed, for every listed model
  • Adds LeWM for training a world model from frames and actions, with a tutorial

Weaknesses

  • Computer vision only; no text or speech pretraining
  • Docker support and one-command pretraining are in the commercial edition
  • Library only: no UI, serving layer or pretrained model hub
  • LeWM requires PyTorch 2.0+ and the timm extra

What it needs

  • Needs PyTorch 2.0+, timm, PyTorch Lightning
  • Models: MoCo, SimCLR, BYOL, SwAV, DINO

What changed

  1. “The API client is removed. If you still need ApiWorkflowClient, pin lightly<1.6 (#1939). These modules and commands are gone:”

From the release notes. No flag means the note does not say so, not that the upgrade is safe.

Also in Training and fine-tuning

See all 68
Also in Training and fine-tuning
RankProjectScore
1peftPython library for LoRA and other parameter-efficient fine-tuning of Transformers models Live demo ↗ (opens in a new tab)21.9k starsApache-2.089 out of 100
2unslothDesktop and web app to run and fine-tune local models Live demo ↗ (opens in a new tab)79.7k starsApache-2.088 out of 100
3trlPost-training library with SFT, DPO, GRPO and KTO trainers19.8k starsApache-2.088 out of 100
#4deepspeedPyTorch library for distributed training of very large models with ZeRO43.3k starsApache-2.087 out of 100
#5megatron-lmDistributed training library for large transformer models on NVIDIA GPUs18.4k starscustom license86 out of 100