Refiners

Refiners can be used in pipelines to improve the quality of the output. They are used to refine the output of a component, and can be used to fix errors, improve formatting, or add additional information.

Using Refiners

Refiners can be used in the CLI with the -r flag. You can specify multiple refiners in a chain, and they will be run in order. The output of one refiner will be used as the input for the next refiner.

janus translate --input janus/cli/ --output janus-translation --llm my-gpt -l python -r ReflectionRefiner -r CodeFormatRefiner

Available Refiners

• FormatRefiner: Base class for formatting refiners.

• CodeFormatRefiner: Refines the formatting of source code so that it can be correctly parsed.

• RequirementsFormatRefiner: Refines the formatting of requirements so that they can be correctly parsed. Recommended for use with generating requirements.

• SimpleRetry: Simple retry refiner that will retry the LLM call a specified number of times.

• ReflectionRefiner: Uses LLM reflection to improve the quality of the output.

• RequirementsReflectionRefiner: Uses LLM reflection to improve the quality of requirements, specifically.

• HallucinationRefiner: Uses LLM reflection to attempt to remove hallucinations from the output.

• FixUMLConnectionsRefiner: Fixes connections in UML diagrams.