A collection of training and benchmarking examples built on LanceDB multimodal data lakehouse to demonstrate how LanceDB performs as the data layer across different model types, and training regimes along with benchmarks and best practices for training with LanceDB.
| Model type | Example |
|---|---|
| Object Detection (AV perception) | object-detection/  |
| ViT (MFU benchmark across backends) | examples/ViT/ |
| VLA (Vision-Language-Action) | examples/lerobot_ray_lance/ |
| World Model / Video Generation | 🚧 |
| VLM | 🚧 |
| LLM | 🚧 |
object-detection/ # AV perception — BDD100K + Geneva + Faster R-CNN
examples/
ViT/ # MFU benchmark: LanceDB vs S3 vs Parquet
lerobot_ray_lance/ # VLA: Ray + LeRobot Diffusion Policy
leWorldModel/ # CogVideo / world model fine-tuning
Each example is self-contained and targets one concrete question.
uv syncSee each example's own README for run instructions.