trainnr/trainnr-artifacts

trainnr research artifacts

The model checkpoints (*.pt) and study logs (*.log) that the trainnr findings records cite. They are not in the code repository: they would add over 100 MB to every clone and plugin install.

Each file sits at the path it has in the repository (docs/artifacts/...), so downloading this dataset into a checkout's root puts every file where a record expects it. MOVED.tsv lists every file with its path, size and SHA-256.

Download

From the root of a trainnr checkout:

uvx --from huggingface_hub hf download trainnr/trainnr-artifacts \
  --repo-type dataset --include "docs/*" --local-dir .

Then check the files against the manifest:

grep -v '^#' docs/artifacts/MOVED.tsv | awk -F'\t' '{print $3"  "$1}' | sha256sum -c

What is here

• docs/artifacts/walk-c1/: the final checkpoint (model_7999.pt) of each of the 15 arms of the microduck walk-c1 study (point, narrow, wide, identified and point-refit, three seeds each).
• docs/artifacts/lift-studies/: the study logs of the SO-101 lift studies.
• docs/artifacts/dagger/: the DAgger round's data generation log.

Every run's identity.json in the repository names its robot, actuator and fit by their stamps.

A .pt file is a Python pickle: load it with torch.load(path, weights_only=True), as trainnr does.

Licence

These files are part of trainnr and are licensed under the Functional Source License, Version 1.1, ALv2 Future License (FSL-1.1-ALv2; see LICENSE): any use other than a competing product or service, and each version also under the Apache License 2.0 two years after it is made available.

The microduck checkpoints were trained on Pollen Robotics' microduck model (Apache-2.0; its meshes are Creative Commons BY-NC-SA) with BAM actuator fits (Rhoban, Apache-2.0). Third-party material is listed in the repository's NOTICE.