Here we use three repositories
- Dataset for training
- A base model for transfer learning of fine-tuning
- Experiment output. Versioned by the timestamp of an experiment.
In this use case, we use
put commands to simplify the commands for programmatic use cases.
- Clone the training code
git clone https://github.com/my-org/my-ml-project.git cd my-ml-project
- Download the dataset and the base model
avc get -o dataset/ s3://email@example.com avc get -o base/ s3://firstname.lastname@example.org
- Train and output your training result (trained model, experiment log, hyperparams, etc) to
- Upload the artifacts
avc put artifacts/ s3://mybucket/experiments/project1@202220303-100504