# AÍLITH — Hugging Face uploader

## Start on Windows

Extract the release ZIP, open `AILITH_HuggingFace_v0.5.0`, and double-click **UPLOAD_TO_HF.bat**.

The launcher needs Python 3.10 or newer. It prepares its own environment, installs the Hugging Face client when needed, and keeps the window open so any error remains visible. The environment lives under your Windows LocalAppData directory, with a temporary-directory fallback.

The uploader then:

1. Finds the release and verifies its SHA-256 checksums.
2. Reuses an existing Hugging Face sign-in or starts the standard login flow.
3. Prompts for the dataset repository. Press Enter to accept the suggested repository under your signed-in account, or enter another account/organization and repository.
4. Asks for public/private visibility when creating a new repository. An existing repository keeps its current visibility.
5. Displays the destination and starts uploading after your confirmation.

Login credentials use Hugging Face's standard cache. The uploader does not write them into the release folder or add them to its upload payload. A token used for publishing needs write access to the destination repository.

## Use the uploader-only download

Extract `AILITH_HuggingFace_Uploader.zip`. You can put `UPLOAD_TO_HF.bat` and `hf_upload.py` into your earlier extracted release folder, next to its `README.md`, or run the launcher from the standalone uploader folder.

When run separately, it asks for the path to your existing release folder **or the original outer release ZIP**. Paste the path, including spaces if present. You do not need to download the research package again.

The earlier package's checksum list still determines its payload. Added uploader files are not uploaded with that earlier package unless they appear in its manifest. The corrected full release includes the uploader in its regenerated manifest.

## Select a repository

Accepted input formats include:

```text
AILITH-Beyond-Reinforcement-Learning
YOUR_NAMESPACE/AILITH-Beyond-Reinforcement-Learning
https://huggingface.co/datasets/YOUR_NAMESPACE/AILITH-Beyond-Reinforcement-Learning
```

A repository name without a namespace uses the signed-in account. For an organization, enter its namespace explicitly and use an account with the appropriate permissions.

The repository receives the release contents at its root, including `README.md`, the manuscript, and the full research ZIP. Matching remote paths are updated. Other remote files are kept. The uploader does not change an existing repository's visibility or delete remote files.

## Offline preview

This command validates the checksums and prints the exact file list without login, SDK import, or network access:

```bash
python hf_upload.py --dry-run --folder "PATH_TO_RELEASE_FOLDER_OR_ZIP"
```

You can also run `UPLOAD_TO_HF.bat --dry-run --folder "PATH_TO_RELEASE_FOLDER_OR_ZIP"` from a Windows terminal. The launcher skips dependency installation in dry-run mode.

## Command-line upload

For Python users, install `huggingface_hub` in a suitable environment, then run:

```bash
python hf_upload.py --folder "PATH_TO_RELEASE_FOLDER_OR_ZIP" --repo YOUR_NAMESPACE/AILITH-Beyond-Reinforcement-Learning
```

For a new repository with visibility chosen explicitly:

```bash
python hf_upload.py --folder "PATH_TO_RELEASE_FOLDER_OR_ZIP" --repo YOUR_NAMESPACE/AILITH-Beyond-Reinforcement-Learning --visibility public
```

`--visibility private` creates a private repository instead. It does not change visibility when the repository already exists. `--yes` skips the final upload prompt and requires an explicit repository; a new or unresolved repository also needs explicit visibility. Sign-in still needs to be available.

## If interrupted or unsuccessful

- For a connection problem or temporary Hub failure, rerun the uploader with the same repository. It retries transient failures and the Hub can reuse already uploaded content.
- For a permission error, use an account or token with write access to that repository. Read-only access is insufficient for publishing.
- For a checksum mismatch, restore the named file from the original download. The uploader stops before sending an altered or incomplete payload.
- If the selected creation visibility differs from the repository's actual visibility, review the reported destination and rerun. Existing repository settings are preserved.
- If Python is missing, install Python 3.10 or newer from [python.org](https://www.python.org/downloads/windows/), including the launcher or PATH option, then reopen the BAT file.
- If package installation fails, check internet access and rerun. Python packages are installed into the uploader environment.

## Verification scope

The focused test suite checks payload selection, checksum failures, ZIP handling, authentication/error behavior, repository destinations, visibility handling, and retries with mocked Hub calls. Offline checks also exercise the complete real release folder and ZIP. SDK signatures are checked against the installed Hugging Face client.

Development tests do not publish to a live repository. The BAT launcher is reviewed as Windows source; native Windows execution is not part of the recorded Linux test environment. The original manuscript and inner research archive remain byte-identical.

Operational references: [Hub uploads](https://huggingface.co/docs/huggingface_hub/guides/upload) and [authentication](https://huggingface.co/docs/huggingface_hub/package_reference/authentication).