code/scripts
| script | purpose | device |
|---|---|---|
fetch_samples.sh |
downloads the public sample data that is not redistributable (sha256-checked) | no |
bench.py |
stage breakdown of warm forwards (host pre split, H2D, each trace, D2H, post, e2e p50 / p99), device-only back-to-back replays, AICLK, agreement with the stored CPU reference; --dispatch / --num-cqs for the D14 matrix (the card / OPT numbers) |
yes, via bin/devrun |
profile_ops.py |
signposted traced replays of image and bev (+ --eager: one eager run with module-stage signposts) under the device profiler (python -m tracy -r -p -v --op-support-count 4000 ...); --graph-report DIR writes a ttnn-visualizer memory / graph report instead |
yes |
profile_summary.py |
per-segment programs / kernel sum / op-to-op gaps / span, per-stage and per-op-code tables and the top ops of a profile_ops.py CSV |
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 |
make_demo.py |
renders media/*.png (BEV / overlay views) from the shipped sample |
yes |
make_goldens.py |
CPU fp32 reference goldens (port form) of the shipped PandaSet sequence, sample0 and nuScenes scene-0103 kf01 into research/bevformer/goldens/ (workspace data; never shipped), the small shipped goldens in tt_bevformer/tests/goldens/ and the smoke reference samples/pandaset_019_f30.reference.json; cross-checks against the research goldens |
no (research venv) |
make_samples.py |
builds tt_bevformer/samples/ and tt_bevformer/calib/pandaset_019.json from the public-data workflow's shippable PandaSet sample |
no |
Everything here ships in code/ on the Hub and inside the image (source.extra_code lists scripts).