tugrulkaya/datalora-lab

DATALORA Lab

DATALORA Lab is a Hugging Face Space concept for showcasing the tugrulkaya/datalora-mistral-7b adapter in a way that is useful before and after live serving is configured.

What the app does

• —offers a polished prompt playground
• —falls back to an offline research demo when no endpoint is attached
• —includes an efficiency explorer for token pruning and top-k expert routing
• —explains the architecture in product-friendly language

Why this design

The adapter itself is interesting, but a 7B LoRA demo is harder to serve reliably on a free CPU Space. This app stays useful on day one and upgrades cleanly once you attach authenticated inference.

Optional Space secrets

• —HF_TOKEN: lets the app call Hugging Face inference with authentication
• —DATALORA_BASE_URL: points the app to a dedicated Inference Endpoint

If neither secret is present, the Space still works in offline demo mode.

Suggested next step

Create the Space on HF as tugrulkaya/datalora-lab, upload these files, and then decide whether you want:

1. 1.a lightweight explainer-only Space
2. 2.a dedicated endpoint-backed live generation Space
3. 3.a GPU Space that loads the model stack directly