OpenKiwi Configuration

OpenKiwi can be configured in a plethora of different ways. In order to allow flexibility in integration into different pipeline OpenKiwi can be used through a CLI (command line interface) or as a python module.

In both cases however, the configuration options can be passed into OpenKiwi in the form of a YAML config file. Either by calling `kiwi.train(config.yaml)` or by passing the YAML on the CLI with `--config`. Below you can find both the instructions on how to interact with OpenKiwi through it’s CLI and an extensive list of the different configuration options accepted by OpenKiwi.

• Command Line Interface
  • Named Arguments
  • Pipelines
  • Sub-commands:
    • train
    • predict
    • jackknife
    • evaluate
• Configuration Options
  • Train Interface
    • NuQE training
    • Predictor training
    • Predictor-Estimator training
    • QUETCH training
    • Linear training
    • training
    • training-optimization
    • training-save-load
  • Predict Interface
    • NuQE predicting
    • Predictor-Estimator predicting
    • QUETCH predicting
    • Linear predicting
    • predicting
  • jackknife Interface
    • jackknifing
  • Evaluate Interface
    • Evaluation of WMT Quality Estimation
  • General Options
    • random
    • gpu
    • I/O
    • save-load
    • Logging