| --- |
| title: Transformers Playground |
| emoji: 🛝 |
| colorFrom: yellow |
| colorTo: gray |
| sdk: gradio |
| sdk_version: 6.30.0 |
| app_file: app.py |
| pinned: false |
| hf_oauth: true |
| hf_oauth_expiration_minutes: 480 |
| hf_oauth_scopes: |
| - jobs |
| - read-repos |
| short_description: Copy official transformers snippets or run them on HF Jobs |
| --- |
| |
| # Transformers Playground |
|
|
| Pick any transformers model on the Hub, get the usage snippets from its |
| `model_doc` page (adapted to your checkpoint), copy them, or run them on |
| Hugging Face Jobs with your own account. |
|
|
| ## How it works |
|
|
| - `indexer.py` sparse-clones `docs/source/en/model_doc` from transformers, |
| extracts the `<hfoptions id="usage">` tabs and usage code blocks, drops |
| training and quantization examples (v1 is inference only), and writes |
| `snippets.json`. |
| - `app.py` matches a model's `config.json` `model_type` to its doc page, |
| swaps the doc's checkpoint for the selected model, adds a generic |
| `pipeline()` snippet from the model's task, and recommends the cheapest |
| Jobs flavor that fits the weights (live prices from `list_jobs_hardware`). |
| - Runs go through `run_job` with the user's OAuth token, so each user pays |
| for their own compute. Jobs need a PRO, Team or Enterprise account. |
|
|
| ## Setup |
|
|
| 1. Create the Space from these files. On first boot it builds the index |
| itself (about 30 seconds). |
| 2. Optional but recommended: publish the index as a dataset and refresh it |
| nightly, then set a Space variable `INDEX_REPO=<you>/transformers-snippets`. |
|
|
| ```bash |
| hf jobs scheduled run @daily python:3.12 --secrets HF_TOKEN -- bash -c \ |
| "pip install -q huggingface_hub && \ |
| curl -sL https://huggingface.co/spaces/<you>/transformers-playground/raw/main/indexer.py -o indexer.py && \ |
| python indexer.py --push <you>/transformers-snippets" |
| ``` |
|
|
