Skip to main content

module cleanlab_codex.types.project

Types for Cleanlab projects. Codex is the API interface to the Cleanlab AI Platform.

class ProjectConfig​

Type representing options that can be configured for a Cleanlab project.

class ProjectConfig(TypedDict):
     clustering_use_llm_matching: 'bool'
     eval_config: 'ConfigEvalConfig'
     llm_matching_model: 'str'
     llm_matching_quality_preset: 'typing.Literal['best', 'high', 'medium', 'low', 'base']'
     lower_llm_match_distance_threshold: 'float'
     max_distance: 'float'
     query_use_llm_matching: 'bool'
     question_match_llm_prompt: 'str'
     question_match_llm_prompt_with_answer: 'str'
     tlm_evals_model: 'str'
     upper_llm_match_distance_threshold: 'float'

property max_distance​

Distance threshold used to determine if two questions in a project are similar. The metric used is cosine distance. Valid threshold values range from 0 (identical vectors) to 1 (orthogonal vectors). While cosine distance can extend to 2 (opposite vectors), we limit this value to 1 since finding matches that are less similar than “unrelated” (orthogonal) content would not improve results of the system querying the Cleanlab project.