Skip to content

README.md

Latest commit

 

History

History
21 lines (15 loc) · 966 Bytes

README.md

File metadata and controls

21 lines (15 loc) · 966 Bytes

Evolution Loop

Demonstrates Adam's self-improving agent via the evolution loop. The program defines an evaluation function that scores LLM output based on length, keyword presence, and structural formatting. It configures adam_evolve with a writing task, an initial strategy, and scoring metrics. The loop iterates, refining the agent's approach until the target score is reached or a plateau is detected, then prints the best output and the final evolved strategy.

Build & Run

# Create a .env file with your API key
echo "ANTHROPIC_API_KEY=sk-ant-..." > .env

make
./evolution

What It Demonstrates

  • Defining a custom evaluation function (adam_evolve_eval_fn)
  • Configuring evolution parameters: task, strategy, metrics, stopping conditions
  • Using adam_evolve_config_defaults and overriding specific fields
  • Progress tracking via adam_evolve_progress_fn
  • Inspecting results: best output, final strategy, stop reason, token usage