Quick start

This short walkthrough checks the installed CLI and shows where a real experiment configuration enters the workflow.

Check the environment

After installing chimera-ml, run:

chimera-ml doctor
chimera-ml --help

doctor reports basic Python, PyTorch, CUDA, MLflow, and registry/plugin information without starting a training run.

Validate a configuration

Install one of the repository's example plugins, then validate a configuration before starting an experiment:

python -m pip install -e examples/va_estimation
chimera-ml validate-config --config-path examples/va_estimation/configs/multimodal_train.yaml

Validation checks the YAML structure and required experiment fields. The example configuration still needs its dataset and output paths configured for an actual training run; see the VA estimation example README for those paths and the subsequent train/eval commands.

Use the Python containers

The smallest library-level interaction uses the public Batch and ModelOutput containers:

import torch

from chimera_ml import Batch, ModelOutput

batch = Batch(
    inputs={"audio": torch.zeros(1, 16000)},
    targets=None,
)
output = ModelOutput(preds=torch.zeros(1, 1))

assert batch.inputs["audio"].shape == (1, 16000)
assert output.preds.shape == (1, 1)

Models consume Batch objects and return ModelOutput objects. The API reference documents these containers and the registry used to connect built-in or plugin components.