dspy-synthetic-data-generation.py

import dspy
from dspy.experimental import SyntheticDataGenerator
from pydantic import BaseModel, Field

llm = dspy.OpenAI(model='gpt-3.5-turbo',api_key=openai_key)

dspy.settings.configure(lm=llm)

class SyntheticFacts(BaseModel):
    fact: str = Field(..., description="a statement")
    veracity: bool = Field(..., description="an assessment of the veracity of the statement")

# Generate synthetic data from pydantic model
generator = SyntheticDataGenerator(schema_class=SyntheticFacts)

examples = generator.generate(sample_size=10)

print(f"Examples from pydantic model: {examples}")

# If you already have some examples and potentially need more:
existing_examples = [dspy.Example({'fact': 'The Earth is round.', 'veracity': 'True'}), 
                     dspy.Example({'fact': 'Water boils at 100°C.', 'veracity': 'True'})]

partial_generator = SyntheticDataGenerator(examples=existing_examples)

augmented_examples = partial_generator.generate(sample_size=8)

print(f"Examples from existing examples: {augmented_examples}")