What’s new
in MLGatee.
Product updates, newest first.
Nightly encrypted backups and daily health checks
The database is backed up every night, encrypted, and kept for 30 days. A daily check watches storage, builds and stuck deployments.
- Uploads stop safely before storage fills; live models keep working
Hugging Face text models
mlgatee.deploy_hf() packs a small text classification or embedding model and its tokenizer into one 8-bit ONNX file. Send plain text to /predict.
- Python client 0.7.0
Keras and PyTorch from your notebook
The Python client converts Keras models to TensorFlow Lite and PyTorch models to ONNX in your own Python, and checks the result against the original before uploading.
- Python client 0.6.0
TensorFlow Lite models
.tflite files with float, float16 or int8 weights. Softmax classifiers also answer /predict_proba. Files that need full TensorFlow are refused with the op's name.
ONNX models
.onnx files are read without loading and checked for custom operators and external data. The Activity log lists each input and output.
Monitoring
Calls, errors, answer times and the last call for the last 24 hours or 7 days, with a By version table. Services report counts and timings only, never inputs.
- Python client: mlgatee usage NAME --window 7d
Roll back to an earlier version
Pick any of your three newest versions. The rollback becomes a new version built from that file, at the same address with the same key.
- Python client: mlgatee.versions() and mlgatee.rollback()
Deploy from your notebook
The Python client deploys a model object with mlgatee.deploy(). Sign in with a personal access token from Settings → Access tokens.
- Tokens expire after 90 days and can be revoked any time
CatBoost, plus XGBoost's own formats
CatBoost .pkl and .joblib files, and XGBoost .json and .ubj files, deploy like any other model.
Replace a model, keep its address and key
Upload a new file for a live model. The current version keeps answering during the build, and stays live if the build fails.