Skip to content

Revert "Merge pull request #68 from jvm123/feat-llm-artifact-side-cha… #82

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Jun 19, 2025
Merged

Revert "Merge pull request #68 from jvm123/feat-llm-artifact-side-cha… #82

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
23 changes: 6 additions & 17 deletions README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -185,7 +185,7 @@ See the [Configuration Guide](configs/default_config.yaml) for a full list of op

## Artifacts Channel

OpenEvolve includes an **artifacts side-channel** that allows evaluators to capture build errors, profiling results, etc. to provide better feedback to the LLM in subsequent generations. This feature enhances the evolution process by giving the LLM context about what went wrong and how to fix it.
OpenEvolve includes a **artifacts side-channel** that allows evaluators to capture build errors, profiling results, etc. to provide better feedback to the LLM in subsequent generations. This feature enhances the evolution process by giving the LLM context about what went wrong and how to fix it.

The artifacts channel operates alongside the traditional fitness metrics.

Expand All @@ -205,28 +205,17 @@ return EvaluationResult(
```

The next generation prompt will include:
```markdown
```
## Last Execution Output
### Stderr
```
SyntaxError: invalid syntax (line 15)

```
### Traceback
```
...
```

## Example: LLM Feedback

An example for an LLM artifact side channel is part of the default evaluation prompt template, which ends with
```markdown
Return your evaluation as a JSON object with the following format:
{{
"readability": [score],
"maintainability": [score],
"efficiency": [score],
"reasoning": "[brief explanation of scores]"
}}
```
The non-float values, in this case the "reasoning" key of the json response that the evaluator LLM generates, will be available within the next generation prompt.

### Configuration

Expand All @@ -251,7 +240,7 @@ export ENABLE_ARTIFACTS=false
### Benefits

- **Faster convergence** - LLMs can see what went wrong and fix it directly
- **Better error handling** - Compilation and runtime failures become learning opportunities
- **Better error handling** - Compilation and runtime failures become learning opportunities
- **Rich debugging context** - Full stack traces and error messages guide improvements
- **Zero overhead** - When disabled, no performance impact on evaluation

Expand Down
44 changes: 11 additions & 33 deletions openevolve/evaluator.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -131,32 +131,16 @@ async def evaluate_program(
eval_result = self._process_evaluation_result(result)

# Add LLM feedback if configured
llm_eval_result = None
if self.config.use_llm_feedback and self.llm_ensemble:
llm_result = await self._llm_evaluate(program_code)
llm_eval_result = self._process_evaluation_result(llm_result)
feedback_metrics = await self._llm_evaluate(program_code)

# Combine metrics
for name, value in llm_result.metrics.items():
for name, value in feedback_metrics.items():
eval_result.metrics[f"llm_{name}"] = value * self.config.llm_feedback_weight

# Store artifacts if enabled and present
if (
artifacts_enabled
and (
eval_result.has_artifacts()
or (llm_eval_result and llm_eval_result.has_artifacts())
)
and program_id
):
self._pending_artifacts[program_id] = {}

# Merge eval_result artifacts with llm artifacts if they exist
if eval_result.has_artifacts():
self._pending_artifacts[program_id].update(eval_result.artifacts)

if llm_eval_result and llm_eval_result.has_artifacts():
self._pending_artifacts[program_id].update(llm_eval_result.artifacts)
if artifacts_enabled and eval_result.has_artifacts() and program_id:
self._pending_artifacts[program_id] = eval_result.artifacts

elapsed = time.time() - start_time
logger.info(
Expand All @@ -172,7 +156,6 @@ async def evaluate_program(
logger.warning(
f"Evaluation attempt {attempt + 1}/{self.config.max_retries + 1} failed for program{program_id_str}: {str(e)}"
)
traceback.print_exc()

# Capture failure artifacts if enabled
if artifacts_enabled and program_id:
Expand Down Expand Up @@ -425,7 +408,6 @@ async def _llm_evaluate(self, program_code: str) -> Dict[str, float]:
json_pattern = r"```json\n(.*?)\n```"
import re

artifacts = {}
avg_metrics = {}
for i, response in enumerate(responses):
json_match = re.search(json_pattern, response, re.DOTALL)
Expand All @@ -444,13 +426,12 @@ async def _llm_evaluate(self, program_code: str) -> Dict[str, float]:
# Parse JSON
result = json.loads(json_str)

# All non-numeric values are artifacts, all numeric values are metrics
metrics = {}
for key, value in result.items():
if not isinstance(value, (int, float)):
artifacts[key] = value
else:
metrics[key] = float(value)
# Filter all non-numeric values
metrics = {
name: float(value)
for name, value in result.items()
if isinstance(value, (int, float))
}

# Weight of the model in the ensemble
weight = self.llm_ensemble.weights[i] if self.llm_ensemble.weights else 1.0
Expand All @@ -462,10 +443,7 @@ async def _llm_evaluate(self, program_code: str) -> Dict[str, float]:
else:
avg_metrics[name] = value * weight

return EvaluationResult(
metrics=avg_metrics,
artifacts=artifacts,
)
return avg_metrics

except Exception as e:
logger.warning(f"Error parsing LLM response: {str(e)}")
Expand Down