Skip to content

speechpro/inex-launcher

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

145 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

GitHub License PyPI - Version PyPI - Python Version

Stars Badge Forks Badge Pull Requests Badge Issues Badge GitHub contributors

InEx Launcher

Lightweight highly configurable Python launcher based on microkernel architecture

Installation

pip install -U inex-launcher

Single-file configuration

inex-launcher lets you describe an entire experiment in a single YAML file. The plugins list defines the initialization order of handlers. For each handler you only need to provide the module path and its options. The execute section describes the final method that will be called.

plugins:
  - loader
  - model
  - trainer

loader:
  module: myproject.data/Loader
  options:
    path: data/train.csv

model:
  module: myproject.nn/Model
  options:
    hidden: 128

trainer:
  module: myproject.training/Trainer
  imports:
    model: plugins.model
    data: plugins.loader
  options:
    epochs: 10

execute:
  method: myproject.training/run
  imports:
    trainer: plugins.trainer

Run this configuration with:

inex config.yaml

You can fine-tune the experiment by merging it with another file using --merge and overriding specific values with --update:

inex config.yaml -m experiment.yaml -u trainer.epochs=20 model.hidden=64

This approach lets you modify parameters or even replace handlers with just a few lines of YAML—no need to edit high‑level Python files.

Key benefits of this approach:

  • High modularity – modules can be swapped directly in the configuration thanks to the plugin-based structure.
  • Clean structure – a clear sequence of plugins followed by a single execute block keeps the file readable.
  • Flexibility – YAML allows complex nested structures, so even advanced scenarios fit in one file.

About

Lightweight highly configurable Python launcher based on microkernel architecture

Resources

License

Stars

Watchers

Forks

Packages

 
 
 

Contributors

Languages