File size: 2,297 Bytes
3bd1680 96e3595 3bd1680 93389f2 4e416cf 3bd1680 96e3595 3bd1680 7b3e625 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 | # Downloading ALTER simulation artifacts
Models: https://huggingface.co/Berkeley-ICON-Lab/ALTER-models
Data and provenance: https://huggingface.co/datasets/Berkeley-ICON-Lab/ALTER-data
Code: https://github.com/labicon/ALTER/tree/e180a013e30d7b7c5478e029229ff9cd3154028e
The simulation code is merged into the public repository’s `main` branch.
The commands below pin the tested release revision for reproducible downloads.
Both artifact uploads are complete. release-manifest.json records immutable
artifact revisions and checksums. bundle-index.json and SHA256SUMS list this
repository’s artifact contents. The manifest pins the commits containing the
artifacts, which precede this documentation commit.
Check out the tested code revision and follow its [installation guide](https://github.com/labicon/ALTER/blob/e180a013e30d7b7c5478e029229ff9cd3154028e/documentation/release/installation.md):
```bash
git clone https://github.com/labicon/ALTER.git
cd ALTER
git checkout e180a013e30d7b7c5478e029229ff9cd3154028e
```
Then, from the repository root run:
```bash
python scripts/download_release.py --manifest release/manifest.json \
--bundles simulation-models simulation-pretraining-data \
simulation-adaptation-data simulation-provenance \
--destination /path/to/ALTER-artifacts
python scripts/materialize_release.py --bundle-root /path/to/ALTER-artifacts \
--output "$PWD/public-validation/inputs"
python scripts/validate_release_models.py --inputs "$PWD/public-validation/inputs" \
--output "$PWD/public-validation/model-check.json" --training-smoke
```
Materialization requires all four bundles. It verifies hashes, relocates portable
artifact references and strictly validates derived checkpoint contracts. Direct
use of portable statistics before materialization is not supported. Only load
trusted pickle/PyTorch files. Short tests check functionality; they do not rerun
full training or establish paper success rates. Hardware v1 is documented separately in [hardware/v1/USAGE.md](hardware/v1/USAGE.md).
## Hardware release code
Use [hardware v1 instructions](hardware/v1/USAGE.md) and the [tested code revision](https://github.com/labicon/ALTER/tree/a4919df35d703fafcf054717f8a53fd876b6a010).
Hardware and simulation use separate download directories and manifests.
|