# code/scripts

| script | purpose | device |
|---|---|---|
| `fetch_samples.sh` | template for downloading public sample data that may not be redistributed (sha256-checked); no source is enabled (nuScenes and METEOR are never fetched by the bundle) | no |
| `make_sample.py` | generates the shipped synthetic sample `samples/synthetic_5cam*`, its preset and its stored CPU reference (`optimise` / `finalise`; research venv, CPU) | no |
| `make_samples.py` | builds the PandaSet 019 f40-43 sample from the research ship sample (default `--dest staging`: the git-ignored `staging_samples_pandaset/`) | no |
| `make_goldens.py` | CPU reference goldens of the PandaSet sample and the research TF records (research venv) | no |
| `bench.py` | stage breakdown of a warm forward, device-only replays, AICLK, agreement checks (the card / OPT numbers) | yes, via `bin/devrun` |
| `profile_ops.py` / `profile_summary.py` | signposted eager + traced frames under the device profiler (`python -m tracy -r -p -v ...`), the ttnn-visualizer report, and their summary | yes / no |
| `run_public_tt.py` | a public sequence (PandaSet 019, nuScenes scene-0103 / 0916 of the tt-models workspace) through the API, scored against the research CPU goldens | yes |
| `make_demo.py` | renders `media/` from the TT outputs (dataset mosaics / BEV plots through the research renderer, TT vs CPU, the synthetic sample) | no |
| `container_smoke.sh` | serve ONE serve profile of the built package (`--profile NAME`; default profile otherwise), run `server/smoke_test.py` against it (asserts ETH dispatch, the 12x10 grid and the profile's pins; compares with the stored CPU reference), keep the evidence (container log, `/info`, the `/predict` output and the result, in `logs/smoke/` of the repo or `--log-dir DIR`), always stop it; one `bin/devrun -t 3600 -k 150` window per profile | yes |

Everything here ships in `code/` on the Hub and inside the image (`source.extra_code` lists `scripts`).