vladimirrotariu/surrealml-vs-onnx-vs-pytorch

We will compare the performance of PyTorch with ONNX and SurrealML, a library designed for uploading and inferring using ML models directly inside a SurrealDB instance.

SurrealML versus PyTorch versus ONNX

SurrealML simplifies storing, loading, and executing trained ML models, working alongside existing frameworks like PyTorch, TensorFlow, scikit-learn, and linfa, while supporting execution in both Python and Rust.

In this repo, you can find the Jupyter notebook behind the following comparison, which shows execution times for SurrealML, as contrasted with PyTorch and ONNX.

To start, clone this repository, and then execute:

poetry install

and then follow the Jupyter notebook attached.

For convenience, you can find here a PDF version of the Jupyter notebook.

System specs

We used the following setup to create the benchmark:

Number of physical CPU cores: 10
Number of logical CPU cores: 20
Total Memory (RAM): 15.47 GB
Operating System: Linux 5.15.153.1-microsoft-standard-WSL2
Processor: x86_64
Python Version: 3.11.10

as well as SurrealDB 1.5.5.