A hypothetical user that is used to send notifications for a company. These users are not main_users (people), but they would have to have ac

Usage: with MaybeSuppressErrors(suppress_errors=True) as suppressor: ... # some potentially error-raising code suppressor

Main trainer class for SGD-based prompt optimization. Implements: - Batch processing with configurable batch size - Dynamic subj

A function caller that keeps track of the returned value, and keeps calling the function until it returns a value that hasn't been returned b

Evaluates model responses using a user-defined reward function. The reward function should take (question, answer1, answer2, standard_an

RAG-based knowledge base for storing and retrieving guidance. Storage format: JSONL with {subject: str, guidance: str, embedding: List[f

Unified client for interacting with user-defined model functions. Supports: - Tuning model (the model being optimized) - Tuner m

Extends LoginManager to also allow Logging out.

Builds system prompts from knowledge base entries. Organizes guidance by subject and creates well-structured system prompts.