Installation and model download
Environment and installation
Create a dedicated Conda environment for LingBot-VLA:
conda create -n lingbotvla python=3.12 -y
conda activate lingbotvla
git clone https://github.com/Robbyant/lingbot-vla.git
cd lingbot-vla
bash install.sh
Download pretrained checkpoints
| Model | Hugging Face | ModelScope | Description |
|---|---|---|---|
| LingBot-VLA-4B | View model | View model | Without depth distillation |
| LingBot-VLA-4B-Depth | View model | View model | Distills 3D geometry from MoGe-2 and LingBot-Depth |
python3 scripts/download_hf_model.py \
--repo_id robbyant/lingbot-vla-4b \
--local_dir lingbot-vla-4b
Checkpoints downloaded before May 1, 2026 may fail to load after the LeRobot v2.1 to v3.0 migration. Download the latest checkpoint if configuration decoding fails.
Prepare base checkpoints
| Checkpoint | Use |
|---|---|
| Qwen2.5-VL-3B-Instruct | Vision-language backbone |
| MoGe-2-vitb-normal | Geometry teacher for the depth-distilled model |
| LingBot-Depth | Depth teacher for the depth-distilled model |
export QWEN25_PATH=/path/to/Qwen2.5-VL-3B-Instruct
Continue to Prepare data or Post-training.