Input validations test the LLM references and/ or user prompts against predefined quality standards. This prevents garbage-in, and garbage-out scenarios, i.e., if you feed poor inputs, you’ll get poor outputs.
Pythia input validators serve the following objectives to protect against unreliable LLM responses:
Validate input data for completeness, accuracy, and relevance.
Reject input data that does not meet the specified criteria.
Log incidents of failed checks for audit and improvement purposes.
Pythia implements the following input validators in each API request: