Zoo Remote Model#

Build FiftyOne remote model zoo integrations that work with register_zoo_model_source, multi-worker DataLoader, and dataset.apply_model.

Install#

curl -sL skil.sh | sh -s -- voxel51/fiftyone-skills

When prompted, select fiftyone-zoo-remote-model from the menu.

Requirements#

FiftyOne
Python 3.9+
PyTorch (or the framework your model uses)

Usage#

Ask your AI assistant:

"Build a remote zoo model wrapper for my Hugging Face detection model"
"Create a FiftyOne zoo source with manifest.json for my custom classifier"
"My zoo model raises ModuleNotFoundError from DataLoader workers — fix it"
"Add VLM-style prompt dispatch to my zoo model"

The skill scaffolds the source layout (manifest.json, __init__.py, zoo.py), wires the model class against TorchModelMixin + SupportsGetItem, and walks the validation steps so the result works under multi-worker DataLoader and dataset.apply_model.

Example#

Register and apply a zoo source written with this skill:

import fiftyone as fo
import fiftyone.zoo as foz

foz.register_zoo_model_source("/path/to/your-zoo-source")
model = foz.load_zoo_model("your-model-name")

dataset = fo.load_dataset("your-dataset")
dataset.apply_model(model, label_field="predictions")

See also#