Instructions to use lowdown-labs/fela-power-grid with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use lowdown-labs/fela-power-grid with Transformers:
# Load model directly from transformers import AutoModel model = AutoModel.from_pretrained("lowdown-labs/fela-power-grid", trust_remote_code=True, dtype="auto") - Notebooks
- Google Colab
- Kaggle
Quickstart: FELA Grid Renewable
A minimal example that loads the forecaster, runs one weather input window, and prints the P10 to P90 power band for the first forecast hour. It runs on CPU; no GPU is required.
Install
Pinned versions, runs from a clean virtual environment:
pip install -r requirements.txt
Get the weights
The safetensors weights ship in the model repo (solar.safetensors, wind.safetensors).
--weights defaults to the track's safetensors file, or point it at a local path.
Run
Solar (default):
python run.py --track solar --weights /path/to/solar.safetensors
Wind:
python run.py --track wind --weights /path/to/wind.safetensors
The script loads with the bundled modeling.load_model (a few line load) and preprocesses
the window with modeling.preprocess_nwp, which standardizes it and validates the shape
(it fails clearly on the wrong shape).
What you should see
The script prints the output shape (1, 99) and, for the center forecast hour,
the P10 (low), P50 (median), and P90 (high) power levels as a fraction of site capacity,
plus the width of the P10 to P90 uncertainty band. Input shapes are solar (1, 6, 20),
wind (1, 12, 15).
The example uses a random input window so it runs without a data download. Replace it with a real NWP window (the same shape) to get a real forecast.