Chimera ML

Chimera ML is a configuration-driven Python framework for building, training, evaluating, and running uni-modal or multi-modal PyTorch experiments. The runtime provides the trainer, registries, callbacks, loggers, and inference pipeline; task-specific components can live in an installed plugin package.

Start here

1. Getting started — install Chimera ML and validate your first experiment configuration.
2. Concepts — understand configuration, registries, BuildContext, and plugins.
3. User guide — configure training, evaluation, logging, inference, sweeps, and CLI workflows.
4. Tutorials — use the real VA, ORAGEN, and affective states example projects.
5. API reference and CLI reference — find precise interfaces and command options.
6. Developer guide — contribute, test, document, and release the project.
7. Releases — read changelog and migration notes.

Core model boundary

A dataloader yields Batch, a model accepts Batch and returns ModelOutput, and losses/metrics consume those objects. Experiments are described in YAML and components are selected by registry key.

Repository examples

The examples are real plugin packages and require their own data or model artifacts:

• VA estimation
• ORAGEN
• Affective states recognition

Install the published package with:

python -m pip install chimera-ml

The current package supports Python >=3.12,<3.13. Continue with Installation.