Pydantic Outputs

Working with Pydantic Models

The completion() function utilizes the Instructor library to provide a quick option to generate structured outputs using Pydantic.

import zyx
from pydantic import BaseModel

class Character(BaseModel):
    superhero : str
    secret_identity : str

zyx.completion(
    "Who is spiderman?",
    model = "ollama/llama3.2",
    response_model = Character
)

---

Using the BaseModel Subclass

zyx has a subclass of Pydantic's BaseModel that has an additional generate() method, specifically built for generating quicker structured outputs with LLMs.

from zyx import BaseModel # Subclass of the Pydantic BaseModel

class Character(BaseModel):
    superhero : str
    secret_identity : str

Character.generate("Who is batman?")

# Character(superhero="Batman", secret_identity="Bruce Wayne")

---

Generating Batch Synthetic Data

You can also generate batch synthetic data using the generate method. Also, instructions are not required when using the generate method. The LLM will be automatically prompted to create synthetic & diverse data.

class SyntheticData(BaseModel):
    food : str
    chemical_compounds : list[str]

SyntheticData.generate(
    n=10,
)

# SyntheticData(
#     food='Apple',
#     chemical_compounds=['Quercetin', 'Cyanidin', 'Chlorogenic acid']
# ),
# SyntheticData(
#     food='Banana',
#     chemical_compounds=['Dopamine', 'Serotonin', 'Catecholamines']
# ),
# SyntheticData(
#     food='Carrot',
#     chemical_compounds=['Beta-carotene', 'Lutein', 'Zeaxanthin']
# ), ....

Chain of Thought with Pydantic Models

from zyx import BaseModel

class LearningPlanWeek(BaseModel):
    tasks : list[str]

class LearningPlan(BaseModel):
    goal : str
    week_1 : LearningPlanWeek
    week_2 : LearningPlanWeek
    week_3 : LearningPlanWeek

LearningPlan.generate(
    "Generate a cohesive 3 week plan for learning python",
    process = "sequential" # This generates each field sequentially, emulating chain of thought
)


# [
#     LearningPlan(
#         goal='Achieve a 20% increase in sales by the end of Q4.',
#         week_1=LearningPlanWeek(
#             tasks=[
#                 'Conduct market research to identify potential customer segments.',
#                 'Develop a targeted marketing campaign to reach new customers.',
#                 'Train the sales team on new sales techniques and product knowledge.',
#                 'Set weekly sales targets and monitor progress.',
#                 'Review and analyze sales data to adjust strategies as needed.'
#             ]
#         ),
#         week_2=LearningPlanWeek(
#             tasks=[
#                 'Implement the marketing campaign and track its effectiveness.',
#                 'Host a webinar to showcase products to potential customers.',
#                 'Follow up with leads generated from the marketing efforts.',
#                 'Conduct a sales team meeting to discuss progress and challenges.',
#                 'Gather feedback from the sales team on customer interactions and adjust
# strategies accordingly.'
#             ]
#         ),
#         week_3=LearningPlanWeek(
#             tasks=[
#                 'Analyze the results of the marketing campaign and sales performance.',
#                 'Identify areas for improvement based on customer feedback and sales
# data.',
#                 'Refine the sales strategy to better target high-potential customer
# segments.',
#                 'Conduct additional training sessions for the sales team based on
# identified gaps.',
#                 'Prepare a report summarizing progress towards the sales goal and next
# steps.'
#             ]
#         )
#     )
# ]