{"id":51060474,"url":"https://github.com/lancedb/training","last_synced_at":"2026-06-23T01:31:08.729Z","repository":{"id":358376778,"uuid":"1156627590","full_name":"lancedb/training","owner":"lancedb","description":"All things training.","archived":false,"fork":false,"pushed_at":"2026-06-01T07:47:38.000Z","size":33905,"stargazers_count":0,"open_issues_count":2,"forks_count":1,"subscribers_count":0,"default_branch":"main","last_synced_at":"2026-06-01T09:25:55.093Z","etag":null,"topics":[],"latest_commit_sha":null,"homepage":null,"language":"Jupyter Notebook","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":null,"status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/lancedb.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":null,"code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null,"notice":null,"maintainers":null,"copyright":null,"agents":null,"dco":null,"cla":null}},"created_at":"2026-02-12T21:37:22.000Z","updated_at":"2026-06-01T07:26:37.000Z","dependencies_parsed_at":null,"dependency_job_id":null,"html_url":"https://github.com/lancedb/training","commit_stats":null,"previous_names":["lancedb/training"],"tags_count":0,"template":false,"template_full_name":null,"purl":"pkg:github/lancedb/training","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/lancedb%2Ftraining","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/lancedb%2Ftraining/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/lancedb%2Ftraining/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/lancedb%2Ftraining/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/lancedb","download_url":"https://codeload.github.com/lancedb/training/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/lancedb%2Ftraining/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":34672250,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-05-26T15:22:16.424Z","status":"online","status_checked_at":"2026-06-22T02:00:06.391Z","response_time":106,"last_error":null,"robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":true,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":[],"created_at":"2026-06-23T01:31:08.042Z","updated_at":"2026-06-23T01:31:08.723Z","avatar_url":"https://github.com/lancedb.png","language":"Jupyter Notebook","funding_links":[],"categories":[],"sub_categories":[],"readme":"A collection of training and benchmarking examples built on [LanceDB](https://lancedb.com) 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.\n\n## Why use LanceDB for training?\n\nOne LanceDB table for the entire training loop\n\n- **Curate \u0026 engineer features** — compute new signals (detections, CLIP embeddings, scene tags, dedup flags) as columns via distributed, checkpointed UDF backfills, then slice with SQL, full-text, and vector search directly on the table. Create a dataloder directly using filtered reads from the table or create a training split as *versioned materialized view*, not a CSV manifest.\n- **Manage at scale** — bytes, metadata, annotations, and embeddings all live in one schema-enforced table, and **zero-copy schema evolution** lets you add a column without rewriting the data: pre-tokenize or pre-embed a multi-TB corpus once and append it as a new column for free.\n- **Load \u0026 train** — With random-access, zero-copy dataloading reads straight from LanceDB tables, from local or object storage, keeping the GPU fed and shards cleanly across Ray workers.\n\n---\n\n### Projects using LanceDB\n\n- **[stable-worldmodel](https://github.com/galilai-group/stable-worldmodel)** — a platform for reproducible world-model research ([paper](https://arxiv.org/abs/2605.21800)) built on a **LanceDB data layer**. **3–4× faster** data loading on Push-T vs HDF5 / MP4 at a fraction of the disk.\n- **[le-wm](https://github.com/lucas-maes/le-wm)** — LeWorldModel, a stable joint-embedding predictive world model from pixels ([paper](https://arxiv.org/abs/2603.19312)), trained on the [stable-worldmodel](https://github.com/galilai-group/stable-worldmodel) platform and its LanceDB data layer.\n- **[lerobot-lancedb](https://github.com/lancedb/lerobot-lancedb)** — a drop-in LanceDB backend for 🤗 LeRobot datasets, referenced in the [official LeRobot docs](https://huggingface.co/docs/lerobot/lerobot-dataset-v3#other-formats-and-implementations). **2–4× faster** data loading across PushT / ALOHA / Koch at identical training quality.\n\n*Building on LanceDB and want to be listed here? Open a PR.*\n\n## Examples\n\n| Model type | Example |\n|---|---|\n| Object Detection (AV perception) | [object-detection/](./object-detection/) \u0026nbsp; [![Blog](https://img.shields.io/badge/blog-read-blue)](https://www.lancedb.com/blog/unifying-the-av-ml-stack-lancedb) |\n| ViT (MFU benchmark across backends) | [examples/ViT/](./examples/ViT/) |\n| VLA (Vision-Language-Action) | [examples/lerobot_ray_lance/](./examples/lerobot_ray_lance/) |\n| World Model / Video Generation |  🚧 |\n| VLM | 🚧  |\n| LLM | 🚧  |\n\n## Repository layout\n\n```\nobject-detection/                 # AV perception — BDD100K + Geneva + Faster R-CNN\nexamples/\n  ViT/                            # MFU benchmark: LanceDB vs S3 vs Parquet\n  lerobot_ray_lance/              # VLA: Ray + LeRobot Diffusion Policy\n  leWorldModel/                   # CogVideo / world model fine-tuning\n```\n\nEach example is self-contained and targets one concrete question.\n\n## Setup\n\n```bash\nuv sync\n```\n\nSee each example's own README for run instructions.\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Flancedb%2Ftraining","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Flancedb%2Ftraining","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Flancedb%2Ftraining/lists"}