#1616th of 23 in Open models
open_clip
PyTorch library for training and running CLIP-style image-text models
- Stars
- 14.2k
- License
- custom license
- Last commit
- Sep 2026
- Last release
- Feb 2026
- Language
- Python
Missed the latest GitHub check; these facts may be older.
Overview
OpenCLIP is an open implementation of OpenAI's CLIP. It loads many pretrained image-text models (LAION-2B, DataComp-1B, DFN, SigLIP, SigLIP2, PE-Core) and trains new ones with DDP or FSDP2. The main branch also adds NaFlex variable-resolution towers, CoCa, MaMMUT and CLAP audio-text training, with breaking changes from the 3.x releases.
Who it is for: ML researchers training or using CLIP-style embedding models
Strengths
- Large zoo of pretrained checkpoints with ImageNet zero-shot results, up to 85.4% in the README table
- Loads OpenAI, SigLIP, SigLIP2, DFN and PE-Core weights through one API
- Training supports DDP, FSDP2, torch.compile strategies and SigLIP loss chunking
- Model cards for the checkpoints are on the Hugging Face Hub
Weaknesses
- Main branch breaks the training API; older scripts need the v3 branch or a 3.x release
- Horovod, torchscript and trace support were removed on main
- Requires torch>=2.6
- Many new model families on main are labelled experimental
What it needs
- GPU optional
- Needs PyTorch >=2.6, timm, Hugging Face Hub
- Models: CLIP, SigLIP, SigLIP2, CoCa, MaMMUT
Also in Open models
See all 23| Rank | Project | Score |
|---|---|---|
| 1 | llmfitRust tool that checks which LLMs fit your RAM, CPU and GPU | 86 out of 100 |
| 2 | layaSingle-pass multilingual classifier for typed choice, score and yes/no decisions | 83 out of 100 |
| 3 | timesfmPretrained time-series foundation model for zero-shot forecasting from Google Research | 74 out of 100 |
| #4 | unlimited-ocrOCR model that parses long documents and multi-page PDFs in one pass | 69 out of 100 |
| #5 | speechPyTorch toolkit for training and deploying ASR, TTS and speech LLM models | 67 out of 100 |