—
GGUF - binary file format for saving models (including LLMs)
General-purpose GPU-optimized Format
—
Saving a model. Time to save a model of size X.
Pickle file. Ideally reload using the same python version.
ONNX. Has a runtime.
Torch Script.
Custom serialize/deserialization script.
→ General problem: saving an object
Serializing data.
Saving / loading any python object. Any object.
Serialization/Deserialization of AI-ML models into JSON objects
- > dill builds on pickle. serializes more complex objects.
Serializing objects and load testing
Using python requests lib and multiprocessing to load test a web server by parallely sending requests.
Python multiprocessing library, natively supported, doesn’t function smoothly inside jupyter NB. This has something to do with serializing/deserializing objects.