""" Structured Output Example ========================= This example demonstrates generating structured data and JSON responses. Main concepts: - Using JSON mode for structured output - Parsing JSON responses reliably - Generating data models and schemas - Extracting structured information from text - Working with response_format parameter """ import json from typing import Dict, Any from kerb.generation import generate, ModelName, GenerationConfig from kerb.parsing import extract_json def example_json_mode_basic(): """Generate JSON output using JSON mode.""" print("\n" + "="*80) print("EXAMPLE 1: Basic JSON Mode") print("="*80) prompt = """Generate a JSON object representing a Python package with the following fields: # %% # Setup and Imports # ----------------- - name: package name - version: semantic version - description: short description - dependencies: list of dependency names""" config = GenerationConfig( model="gpt-4o-mini", response_format={"type": "json_object"}, temperature=0.7 ) print("\nPrompt: Generate a Python package info JSON\n") response = generate(prompt, config=config) print("Raw Response:") print(response.content) # Parse JSON try: data = json.loads(response.content) print("\nParsed JSON:") print(json.dumps(data, indent=2)) except json.JSONDecodeError as e: print(f"JSON parse error: {e}") def example_structured_data_extraction(): """Extract structured data from unstructured text.""" print("\n" + "="*80) print("EXAMPLE 2: Structured Data Extraction") print("="*80) text = """ Python 3.11 was released on October 24, 2022. It includes several performance improvements and new features like exception groups and the tomllib module. """ prompt = f"""Extract the following information from the text and return as JSON: - language: programming language name - version: version number - release_date: release date - features: list of mentioned features Text: {text} Return only valid JSON.""" config = GenerationConfig( model="gpt-4o-mini", response_format={"type": "json_object"}, temperature=0.3 ) response = generate(prompt, config=config) print("Source Text:") print(text.strip()) print("\nExtracted Data:") data = extract_json(response.content) if data: print(json.dumps(data, indent=2)) else: print("Failed to extract JSON") # %% # Example Schema Generation # ------------------------- def example_schema_generation(): """Generate data that follows a specific schema.""" print("\n" + "="*80) print("EXAMPLE 3: Schema-Based Generation") print("="*80) schema = { "type": "object", "properties": { "function_name": {"type": "string"}, "parameters": { "type": "array", "items": { "type": "object", "properties": { "name": {"type": "string"}, "type": {"type": "string"}, "description": {"type": "string"} } } }, "return_type": {"type": "string"}, "description": {"type": "string"} } } prompt = f"""Generate a JSON object for a Python function that calculates the factorial of a number. Follow this schema exactly: {json.dumps(schema, indent=2)} Return only valid JSON.""" config = GenerationConfig( model="gpt-4o-mini", response_format={"type": "json_object"} ) response = generate(prompt, config=config) print("Generated Function Specification:") data = extract_json(response.content) if data: print(json.dumps(data, indent=2)) def example_multiple_entities(): """Extract multiple structured entities from text.""" print("\n" + "="*80) print("EXAMPLE 4: Multiple Entity Extraction") print("="*80) text = """ FastAPI is a modern web framework for Python. It was created by Sebastian Ramirez. Django is a high-level Python web framework created by Adrian Holovaty and Simon Willison. Flask is a micro web framework written in Python, created by Armin Ronacher. """ prompt = f"""Extract information about each framework mentioned in the text. Return a JSON object with a 'frameworks' key containing an array of objects. Each object should have: name, type, language, and creators (array). Text: {text} Return only valid JSON.""" config = GenerationConfig( model="gpt-4o-mini", response_format={"type": "json_object"}, temperature=0.2 ) response = generate(prompt, config=config) print("Extracted Frameworks:") data = extract_json(response.content) if data and "frameworks" in data: for framework in data["frameworks"]: print(f"\n {framework.get('name', 'Unknown')}:") print(f" Type: {framework.get('type', 'N/A')}") print(f" Language: {framework.get('language', 'N/A')}") print(f" Creators: {', '.join(framework.get('creators', []))}") # %% # Example Validation And Correction # --------------------------------- def example_validation_and_correction(): """Generate and validate structured output.""" print("\n" + "="*80) print("EXAMPLE 5: JSON Validation and Correction") print("="*80) prompt = """Generate a JSON object representing a user profile with: - username (string, lowercase, no spaces) - email (valid email format) - age (integer, 18-100) - interests (array of strings) - created_at (ISO 8601 timestamp) Make up realistic data. Return only valid JSON.""" config = GenerationConfig( model="gpt-4o-mini", response_format={"type": "json_object"} ) response = generate(prompt, config=config) print("Generated User Profile:") try: data = json.loads(response.content) # Validate is_valid = True issues = [] if not isinstance(data.get("username"), str) or " " in data.get("username", ""): issues.append("Username contains spaces") is_valid = False if not isinstance(data.get("age"), int) or not (18 <= data.get("age", 0) <= 100): issues.append("Age out of range") is_valid = False if not isinstance(data.get("interests"), list): issues.append("Interests not a list") is_valid = False print(json.dumps(data, indent=2)) print(f"\nValidation: {'PASSED' if is_valid else 'FAILED'}") if issues: print("Issues found:") for issue in issues: print(f" - {issue}") except json.JSONDecodeError as e: print(f"Invalid JSON: {e}") def example_code_to_json(): """Convert code into structured JSON representation.""" print("\n" + "="*80) print("EXAMPLE 6: Code to JSON Conversion") print("="*80) code = """ def calculate_bmi(weight: float, height: float) -> float: '''Calculate Body Mass Index. Args: weight: Weight in kilograms height: Height in meters Returns: BMI value ''' return weight / (height ** 2) """ prompt = f"""Analyze this Python function and convert it to JSON with: - name: function name - docstring: the docstring text - parameters: array of {{name, type, description}} - return_type: return type annotation - body: brief description of what it does Function: {code} Return only valid JSON.""" config = GenerationConfig( model="gpt-4o-mini", response_format={"type": "json_object"} ) response = generate(prompt, config=config) print("Original Code:") print(code) print("\nJSON Representation:") data = extract_json(response.content) if data: print(json.dumps(data, indent=2)) # %% # Main # ---- def main(): """Run all structured output examples.""" print("\n" + "#"*80) print("# STRUCTURED OUTPUT EXAMPLES") print("#"*80) try: example_json_mode_basic() except Exception as e: print(f"\nExample 1 Error: {e}") try: example_structured_data_extraction() except Exception as e: print(f"\nExample 2 Error: {e}") try: example_schema_generation() except Exception as e: print(f"\nExample 3 Error: {e}") try: example_multiple_entities() except Exception as e: print(f"\nExample 4 Error: {e}") try: example_validation_and_correction() except Exception as e: print(f"\nExample 5 Error: {e}") try: example_code_to_json() except Exception as e: print(f"\nExample 6 Error: {e}") print("\n" + "#"*80) print("# Examples completed") print("#"*80 + "\n") if __name__ == "__main__": main()