Welcome to nnbench, a framework for benchmarking machine learning models. The main goals of this project are

1. To provide a portable, easy-to-use solution for model evaluation that leads to better ML experiment organization, and
2. To integrate with experiment and metadata tracking solutions for easy adoption.

On a high level, you can think of nnbench as "pytest for ML models" - you define benchmarks similarly to test cases, collect them, and selectively run them based on model type, markers, and environment info.

What's new is that upon completion, you can stream the resulting data to any sink of your choice (including multiple at the same), which allows easy integration with experiment trackers and metadata stores.

See the quickstart for a lightning-quick demo, or the examples for more advanced usages.

⚠️ nnbench is an experimental project - expect bugs and sharp edges.

Install it directly from source, for example either using pip or poetry:

pip install nnbench
# or
poetry add nnbench

To understand how nnbench works, you can run the following in your Python interpreter:

# example.py
import nnbench


@nnbench.benchmark
def product(a: int, b: int) -> int:
    return a * b


@nnbench.benchmark
def power(a: int, b: int) -> int:
    return a ** b


runner = nnbench.BenchmarkRunner()
reporter = nnbench.ConsoleReporter()
# run the above benchmarks with the parameters `a=2, b=10`...
record = runner.run("__main__", params={"a": 2, "b": 10})
reporter.display(record)  # ...and print the results to the terminal.

# results in a table look like the following:
# ┏━━━━━━━━━━━┳━━━━━━━┳━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━┓
# ┃ Benchmark ┃ Value ┃ Wall time (ns) ┃ Parameters        ┃
# ┡━━━━━━━━━━━╇━━━━━━━╇━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━┩
# │ product   │ 20    │ 1917           │ {'a': 2, 'b': 10} │
# │ power     │ 1024  │ 583            │ {'a': 2, 'b': 10} │
# └───────────┴───────┴────────────────┴───────────────────┘

Watch the following video for a high level overview of the capabilities and inner workings of nnbench.

For a more realistic example of how to evaluate a trained model with a benchmark suite, check the Quickstart. For even more advanced usages of the library, you can check out the Examples in the documentation.

We encourage and welcome contributions from the community to enhance the project. Please check discussions or raise an issue on GitHub for any problems you encounter with the library.

For information on the general development workflow, see the contribution guide.

The nnbench library is distributed under the Apache-2 license.