GitHub - carbon-re/skeights: Serialize fitted scikit-learn models to safetensors + JSON. No pickle.

Serialize fitted scikit-learn models to safetensors + JSON.

No pickle. No joblib. Just weights and config.

skeights saves a fitted model as two files: a .json file containing hyperparameters and structural config, and a .safetensors file containing the numeric arrays (coefficients, tree splits, leaf values). The JSON is human-readable, so you can inspect, grep, and diff model config without loading it. The safetensors format is compact, typed, and memory-mappable. Neither file executes arbitrary code on load, so loading untrusted models is safe.

Install

Quick start

import skeights
from sklearn.linear_model import Ridge
from sklearn.pipeline import Pipeline
from sklearn.preprocessing import StandardScaler

pipe = Pipeline([
    ("scaler", StandardScaler()),
    ("model", Ridge(alpha=0.1)),
])
pipe.fit(X_train, y_train)

# Save
skeights.save(pipe, "model.safetensors", "model.json")

# Load and predict
loaded = skeights.load("model.safetensors", "model.json")
predictions = loaded.predict(X_test)

Documentation

Full docs, API reference, and supported estimators: carbon-re.github.io/skeights

License

MIT