# Runtime data

The reproducible production corpus, FAISS indexes, and ONNX models are mirrored
in this repository through Git Large File Storage (Git LFS). A normal Git clone
stores only small pointer files until Git LFS downloads the corresponding
artifacts.

The deployed Docker Space at
[`NoiwaDEX/SvaraSetu`](https://huggingface.co/spaces/NoiwaDEX/SvaraSetu)
downloads a pinned
[`NoiwaDEX/SvaraSetu-assets`](https://huggingface.co/NoiwaDEX/SvaraSetu-assets)
bundle containing the same prebuilt corpora, indexes, and models. Runtime
caches, evaluation audio, local user data, and provider credentials are never
published.

Install Git LFS before cloning to receive the mirrored assets automatically:

```powershell
git lfs install
git clone https://github.com/SahilRakhaiya05/SvaraSetu.git
```

Alternatively, prepare the assets with the ingestion utilities under
`svarasetu/ingest/`, download the pinned Hugging Face bundle, or mount prebuilt
artifacts into `data/` when deploying.
The expected local directories are:

```text
data/
├── corpus/
├── indexes/
├── models/
├── cache/
├── eval/
└── raw/
```

Do not add provider credentials, runtime caches, evaluation audio, or
`corpus/svarasetu_user.jsonl`. Model and index binaries must remain under the
existing Git LFS rules in `.gitattributes`.