LLM Interaction

This section describes how Cadence integrates with large language models (LLMs) to generate code mutations for optimization tasks.

Overview

• How prompts are constructed with src.prompt_sampler.build().
• Supported LLM backends and configuration.

Usage

Example:

from src.tasks.tsp_task import TSPTask
from src.prompt_sampler import build
from src.llm import generate

task = TSPTask(n_cities=10)
cities = task.generate_inputs(seed=0)
prompt = build(task, cities)
mutations = generate(prompt)

Configuration

Specify API keys and model settings via environment variables (e.g., GENAI_API_KEY).