Initialization
dspy.configure()
Configure the global settings for DSPy.
Python
dspy.configure(lm=lm, rm=rm)Signatures
dspy.Signature
Define the input/output structure of a task.
Python
class QA(dspy.Signature):
"""Answer questions."""
question = dspy.InputField()
answer = dspy.OutputField()Inline Signatures: "question -> answer"
Modules
dspy.Predict
Basic module for using a signature.
pred = dspy.Predict("input -> output")dspy.ChainOfThought
Adds step-by-step reasoning.
cot = dspy.ChainOfThought("input -> output")dspy.ReAct
Agent loop for reasoning and acting.
agent = dspy.ReAct(signature, tools=[...])Optimizers
dspy.BootstrapFewShot
Optimizes by selecting effective few-shot examples.
opt = dspy.BootstrapFewShot(metric=metric_fn)dspy.MIPRO
Optimizes instructions and examples.
opt = dspy.MIPRO(metric=metric_fn)Evaluation
dspy.Evaluate
Run evaluation on a dataset.
Python
evaluator = dspy.Evaluate(devset=devset, metric=metric_fn)
score = evaluator(program)