| title: Embedding Fleet | |
| emoji: 🛰️ | |
| colorFrom: blue | |
| colorTo: gray | |
| sdk: gradio | |
| sdk_version: "5.49.1" | |
| app_file: app.py | |
| pinned: false | |
| # Embedding Fleet — Jobs control plane | |
| Run-level dashboard for a fan-out embedding run launched with | |
| [`launch-embedding-fleet.py`](https://huggingface.co/datasets/uv-scripts/embeddings/blob/main/launch-embedding-fleet.py) | |
| from [uv-scripts/embeddings](https://huggingface.co/datasets/uv-scripts/embeddings): | |
| documents-processed progress, tokens, live ~$ cost vs the hard timeout ceiling, ETA, | |
| GPU utilization, replica health, and a per-worker table. | |
| Data sources: the run bucket (manifest + worker heartbeats, polled) and the Jobs API | |
| (stages, durations, one metrics sample per running job per tick). | |
| Config (Space variables): `FLEET_BUCKET` (default `davanstrien/embedding-runs`), | |
| `FLEET_NAMESPACE`, `FLEET_POLL_SECS`. Secret: `HF_TOKEN` with read access to jobs + | |
| the bucket. Deep-link a run with `?run=<run-id>`. | |