Entry point: OK Transformer website
A collection of self-contained notebooks on machine learning theory, engineering, and operations. I try to cover topics that frequently come up as building blocks for applications or further theory. I also explore areas where I want to clarify my understanding or delve into details that I personally find interesting or intriguing.
The notebooks should ideally run end-to-end with reproducible results between runs. Exact output values may change due to external dependencies such as differences with hardware and dataset versions, or implementation quirks like nondeterminism, but the conclusions should still generally hold. Please open an issue if you find otherwise 👀 (as I often do)!
git clone git@github.com:particle1331/ok-transformer.git && cd ok-transformer
pip install -U tox
tox -e build
The notebooks can be found in docs/nb.
To run them, create a venv using pdm:
pip install -U pdm
pdm install
Use the resulting .venv as Jupyter kernel.
The following libraries will be installed (see pdm.lock):
╭────────────────────────────────────┬────────────────╮
│ fastapi │ 0.111.0 │
│ Flask │ 3.0.3 │
│ keras │ 2.15.0 │
│ lightning │ 2.3.0 │
│ matplotlib │ 3.9.0 │
│ mlflow │ 2.13.2 │
│ numpy │ 1.26.4 │
│ optuna │ 3.6.1 │
│ pandas │ 2.2.2 │
│ scikit-learn │ 1.5.0 │
│ scipy │ 1.13.1 │
│ seaborn │ 0.13.2 │
│ tensorflow │ 2.15.1 │
│ tensorflow-datasets │ 4.9.6 │
│ tensorflow-estimator │ 2.15.0 │
│ torch │ 2.2.2 │
│ torchaudio │ 2.2.2 │
│ torchinfo │ 1.8.0 │
│ torchmetrics │ 1.4.0.post0 │
│ torchvision │ 0.17.2 │
│ uvicorn │ 0.30.1 │
│ xgboost │ 2.0.3 │
╰────────────────────────────────────┴────────────────╯
GPU 0: Tesla P100-PCIE-16GB
CPU: Intel(R) Xeon(R) CPU @ 2.00GHz
Core: 1
Threads per core: 2
L3 cache: 38.5 MiB
Memory: 15 Gb