Description
Bloom is an open-source tool designed for automated behavior assessment in Language Learning Models (LLMs). Functioning as a structured evaluation system, Bloom uses an evaluation configuration, or 'seed', which details the desired behavior, model transcripts, and the interaction types to analyze.
The tool then creates a series of interactions with the target model, intended to reveal the selected behavior. The growth of the evaluation series depends on its initial setup, setting it apart from other evaluations that might use a consistent elicitation approach and prompting style. This enhances the distinctiveness and adaptability of each evaluation.
The tool also enables users to incorporate API keys from providers and to configure behavioral evaluation elements through behaviors.json and seed.yaml files.
Bloom also features an interactive viewer, offering an easy-to-use interface for exploring transcripts from the run and displaying conversation flows with appropriate formatting.
Pricing Plans
Releases
Reviews
Pros & Cons
Pros
Open-source
Automated assessment
Distinctive evaluation
Cons
Complex initial configuration
Dependent on evaluation 'seed'
API keys might be needed
