PyTorch-Tempest

This repository has my pipeline for training neural nets.

Main frameworks used:

• hydra
• pytorch-lightning

The main ideas of the pipeline:

• all parameters and modules are defined in configs;
• prepare configs beforehand for different optimizers/schedulers and so on, so it is easy to switch between them;
• have templates for different deep learning tasks. Currently, image classification and named entity recognition are supported;

Setup

Recommended (with uv):

brew install uv          # or see https://docs.astral.sh/uv/getting-started/installation/
uv sync                  # creates .venv from pyproject.toml + uv.lock

Then prefix commands with uv run, e.g. uv run python train.py ..., or source .venv/bin/activate to use the env directly.

Alternative (pip):

python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt

Running

This will run training on MNIST (data will be downloaded):

uv run python train.py --config-name mnist_config model.encoder.params.to_one_channel=True

Running on MPS (M1 macbook)

uv run python train.py --config-name mnist_config model.encoder.params.to_one_channel=True trainer.accelerator=mps +trainer.devices=1 optimizer=adan training.lr=0.001

Running on MPS (M1 macbook) with schedule free optimizer https://github.com/facebookresearch/schedule_free/tree/main

uv run python train.py --config-name mnist_config model.encoder.params.to_one_channel=True trainer.accelerator=mps trainer.devices=1 optimizer=adamwschedulefree training.lr=0.001 scheduler.params.patience=100

The default run:

uv run python train.py

The default version of the pipeline is run on imagenette dataset. To do it, download the data from this repository: https://github.com/fastai/imagenette unzip it and define the path to it in conf/datamodule/image_classification.yaml path