Generate structured data

from guardrails import Guard
from guardrails.hub import RegexMatch
from pydantic import BaseModel, Field
from typing import List

NAME_REGEX = "^[A-Z][a-z]+\s[A-Z][a-z]+$"

class Delivery(BaseModel):
    custome_name: str=Field(validators=[RegexMatch(regex=NAME_REGEX)], description="customer name")
    pickup_time: str=Field(description="date and time of pickup")
    pickup_location: str=Field(description="address of pickup")
    dropoff_time: str=Field(description="date and time of dropoff")
    dropoff_location: str=Field(description="address of dropoff")
    price: str = Field(description="price of delivery with currency symbol included")

class Schedule(BaseModel):
    deliveries: List[Delivery]

guard = Guard.for_pydantic(Schedule)
chat_history="""
nelson and murdock: i need a pickup 797 9th Avenue, manila envelope, June 3 10:00am with dropoff 10:30am Courthouse, 61 Center Street C/O frank james
operator: quote - $23.00
neslon and murdock: perfect, we accept the quote
operator: 797 9th ave, 10:00am pickup comfirmed
abc flowers: i need a pickup of a flowers from abc flowers at 21 3rd street at 11:00am on june 2 with a dropoff at 75th Ave at 5:30pm same day
operator: 21 3rd street flowers quote - $14.50
abc flowers: accepted
polk and wardell: i need a pickup of a bagels from Bakers Co at 331 5th street at 11:00am on june 3 with a dropoff at 75th Ave at 5:30pm same day
operator: 331 5th street bagels quote - $34.50
polk and wardell: accepted
"""

prompt = """
From the chat exchanges below extract a schedule of deliveries.
Chats:
${chat_history}
"""

messages = [{
  "role": "system",
  "content": "You are a helpful assistant."
}, {
  "role": "user",
  "content": prompt
}]

tools = [] # an open ai compatible list of tools

response = guard(
    model="gpt-4o",
    messages=messages,
    prompt_params={"chat_history": chat_history},
    tools=guard.json_function_calling_tool(tools),
    tool_choice="required",
)

prompt+="""

${gr.complete_json_suffix_v3}
"""
response = guard(
    model="gpt-4o",
    messages=[messages[0],{
      "role": "user",
      "content": prompt
    }],
    prompt_params={"chat_history": chat_history},
)

g = Guard.for_pydantic(NewFriends, output_formatter="jsonformer")

# JSONFormer is only compatible with HF Pipelines and HF Models:
from transformers import pipeline
pipe = pipeline("text-generation", "TinyLlama/TinyLlama-1.1B-Chat-v1.0")

# Inference is straightforward:
out = g(pipe, messages=[{"role":"user","content":prompt}]).validated_output

# `out` is a dict.  Format it as JSON for readability:
import json
print(json.dumps(out, indent=2))

messages[-1]["content"]+="""

${gr.complete_json_suffix_v3}
"""
response = guard(
    model="gpt-4o",
    messages=messages,
    prompt_params={"chat_history": chat_history},
    response_format={ "type": "json_object" }
)

response = guard(
    model="gpt-4o",
    messages=messages,
    prompt_params={"chat_history": chat_history},
    response_format=g.response_format_json_schema()
)