[Question]LLM prompt evolution optimisation with feedback

[filbert-c]

Hi, i have been looking through the various examples in the llm prompt optimisation. I am trying to figure out how to optimise my own prompt by testing with the tutorial code. I notice

openevolve/examples/llm_prompt_optimization/config_qwen3_evolution.yaml

Line 66 in ba1ca44

use_llm_feedback: true

using llm_feedback is set to true which supposedly improves the performance. However, unlike circle with artifacts example, it seems the evaluator.py in llm optimisation does not have the artifact code option available.

openevolve/examples/circle_packing_with_artifacts/evaluator.py

Line 192 in ba1ca44

def evaluate(program_path):

So if it is not implemented, this option cant be used ? If i understand correctly, the feedback provides incorrect examples to the LLM optimiser based on the current prompt.

[codelion]

The artifacts side-channel was meant to pass on program errors during evaluation so that the llm can correct them. With the prompt optimizer example we are not generating code but a new prompt so it is not used in the evaluator.