DSPy-Mipro

Methods

__init__

It shares the same parameters as the Trainer class, with some additional unique parameters.

unique parameters:

• num_candidates: Number of candidate prompts to generate.
• max_steps: Maximum number of optimization steps.
• minibatch_size: Size of minibatches for evaluation.
• max_bootstrapped_demos: Maximum number of bootstrapped examples. (examples with generated internal steps)
• max_labeled_demos: Maximum number of labeled examples. (examples selected from trainset)
• success_score: score threshold to determine if model is successful to solve the task.