LeVJEPA

LeVJEPA produces clip embeddings and spatial patch tokens from video.

任务: embeddings
尺寸: l at 224 px
安装: pip install libreyolo
支持层级: 仅推理，自 v1.6.0 起。仅支持预测、验证和导出。训练相关的功能不适用。
上游: LeVJEPA，由 MLO Lab 发布，采用 MIT code; CC-BY-NC-4.0 weights 许可。论文、源码
许可: 代码采用 MIT，权重采用 MIT code; CC-BY-NC-4.0 weights。商用

Install

bash

pip install "libreyolo"

Predict

Python

from libreyolo import LibreYOLO, SAMPLE_IMAGE model = LibreYOLO("LibreLeVJEPAl-embed.pt", device="cpu")result = model(SAMPLE_IMAGE)print(result.embeddings)print(model.embed_tokens(SAMPLE_IMAGE).shape)

The encoder uses 16 frames at 224 pixels. Finite videos use a centered window sampled at approximately 7.5 FPS. embed_tokens() exposes patch tokens. Training is not supported. The TorchScript graph requires direct clip input with batch 1. Pretrained weights are CC-BY-NC-4.0.

Export

任务	ONNX	TorchScript	ExecuTorch	TensorRT	OpenVINO	Paddle	MNN	RKNN	ncnn	TFLite	CoreML	Core AI
Embeddings	Embeddings to ONNX：支持	Embeddings to TorchScript：支持	Embeddings to ExecuTorch：支持	Embeddings to TensorRT：支持	Embeddings to OpenVINO：支持	Embeddings to Paddle：不支持	Embeddings to MNN：不支持	Embeddings to RKNN：不支持	Embeddings to ncnn：不支持	Embeddings to TFLite：不支持	Embeddings to CoreML：不支持	Embeddings to Core AI：不支持

Python

from libreyolo import LibreYOLO, SAMPLE_IMAGE model = LibreYOLO("LibreLeVJEPAl-embed.pt", device="cpu")model.export(format="onnx")

Export setup lists format dependencies and loading exported artifacts.

Checkpoints

文件	权重许可
Embeddings
LibreLeVJEPAl-embed.pt	cc-by-nc-4.0

上面的每个文件目前都在 LibreYOLO 组织中，并会在首次使用时下载。

Licensing

请检查你所下载的具体权重在 Hugging Face 仓库中的许可。LibreYOLO 组织里的每个检查点都附有许可，同一家族内也不一定相同。该仓库是权威来源；以下摘要说明本页上次验证时适用的情况。

这里只说明涉及的许可证，不构成法律意见。如果答案对商用很重要，请自行阅读许可证并咨询法律顾问。

原始工作: LeVJEPA, MLO Lab
上游许可: MIT code; CC-BY-NC-4.0 weights
上游来源: github.com/MLO-lab/LeVJEPA
LibreYOLO 代码: MIT
权重: 采用 MIT code; CC-BY-NC-4.0 weights 许可，重新发布在 huggingface.co/LibreYOLO
解读: The repository declares an explicit CC-BY-NC-4.0 exception for pretrained weights. Commercial use of those weights is excluded.

Citation

@misc{kuhn2026levjepaefficientscalable,
      title={LeVJEPA: Efficient & Scalable Video Pretraining without the Heuristics}, 
      author={Lukas Kuhn and Lucas Maes and Giuseppe Serra and Quentin Le Lidec and Yann LeCun and Randall Balestriero and Florian Buettner},
      year={2026},
      eprint={2608.27395},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2608.27395}, 
}

复制自作者在 raw.githubusercontent.com/MLO-lab/LeVJEPA/main/README.md 上提供的引用块。