The model has 3B active parameters. We put the code, homepage, paper and model links here:

- Code: https://github.com/bytedance/Lance

- Homepage: https://lance-project.github.io/

- Paper: https://arxiv.org/abs/2605.18678

- Model: https://huggingface.co/bytedance-research/Lance

p.s. Lance is a research project, not a polished product. The model was trained using fewer than 128 GPUs.