UPLOAD_GUIDE.md

Uploading the prepared Hugging Face release

Recommended destination: PureOne/Avenyra-Geometric-Memory-Physics, repository type dataset. This is a suggested identifier based on the supplied namespace, not a claim that this repository already exists or has been uploaded. Use the namespace and repository name you actually control.

Web upload

1. Extract the release ZIP. Open the extracted Avenyra_HuggingFace_Release_v3.0.0 folder.
2. In your Hugging Face account, create or select a dataset repository. Choose its visibility yourself.
3. Upload the contents of the extracted folder to the repository root. README.md, data/, research/ and the guides must remain at their declared relative paths. Uploading only the ZIP gives a downloadable archive but does not expose the configured JSONL files to the dataset viewer.
4. Review the rendered dataset card and the seven configuration names. Confirm each has the declared test split and row count in DATA_SCHEMA.json.
5. Check the PDF and JSONL downloads, and record the resulting commit identifier. Use that immutable revision when citing or loading the data.

The source package does not select a reuse license. Read LICENSE_STATUS.md; the prepared card intentionally omits a license tag rather than assigning terms on the author's behalf.

CLI upload

The official CLI is provided by huggingface_hub. Authenticate locally with your own account; the release contains no token. From the extracted release root:

python -m pip install -U huggingface_hub
hf auth login
python tools/upload_release.py --repo-id PureOne/Avenyra-Geometric-Memory-Physics

The last command is a dry run: it prints the dataset destination and intended command without contacting the Hub. When ready to publish:

python tools/upload_release.py --repo-id PureOne/Avenyra-Geometric-Memory-Physics --upload

The uploader uses the locally installed CLI, detects its supported dataset-type flag from hf upload --help, and invokes it with an argument list. It refuses repository identifiers outside the namespace/name form. It uploads the validated release root and never asks for a token on the command line. The actual operation creates or changes the specified remote repository according to the CLI and your account permissions; inspect your destination and visibility before executing it.

For manual CLI use, the current official documentation gives this form:

hf upload PureOne/Avenyra-Geometric-Memory-Physics . . --repo-type dataset

Some CLI releases expose the shorter --type flag; the included uploader adapts to either documented help form. See the official CLI guide and dataset configuration guide.

Windows launcher

UPLOAD_TO_HUGGINGFACE.bat runs the dry-run plan by default and pauses before closing. Run it from a terminal with the repository identifier and optional --upload argument:

UPLOAD_TO_HUGGINGFACE.bat PureOne/Avenyra-Geometric-Memory-Physics
UPLOAD_TO_HUGGINGFACE.bat PureOne/Avenyra-Geometric-Memory-Physics --upload

Install Python and the CLI and authenticate first. The launcher retains a nonzero failure exit code and keeps the final output visible. It does not install packages, authenticate, or upload unless explicitly invoked with --upload.

After upload

Record the repository commit; run the optional loader with that revision; compare each configuration's row count; and inspect the rendered card and PDF. The card's test label is an organizational split, not a held-out benchmark. No custom remote dataset code is needed. Uploading prepares a research corpus for review; it does not change the scientific status of any claim.