Sample a parent from the archive, preferring programs from the specified island Args: island_id: The island to prefer pr

Sample a completely random parent from a specific island (uniform distribution) Args: island_id: The island to sample fr

Sample inspiration programs for the next evolution step. For proper island-based evolution, inspirations are sampled ONLY from the

Sample a parent program from the current island for the next evolution step Returns: Parent program from current island

Save the best program Args: program: Best program (if None, uses the tracked best program)

Save a checkpoint Args: iteration: Current iteration number

Helper for min-max scaling

Serialize config object to a dictionary that can be pickled

Serialize feature_stats for JSON storage Returns: Dictionary that can be JSON-serialized

Set up logging

Update the archive of elite programs Args: program: Program to consider for archive

Update the absolute best program tracking Args: program: Program to consider as the new best

Update the reference set for diversity calculation

Update the best program tracking for a specific island Args: program: Program to consider as the new best for the island

Update running statistics

Validate cascade evaluation configuration and warn about potential issues Args: module: The loaded evaluation module

Validate migration didn't create inconsistencies Checks that: 1. Program island metadata matches actual island assignment

Write an artifact to a file

Append a single trace to a JSONL file, useful for streaming data. Args: trace: A single trace object. output_path: The path

Create and track a task that will run in the pool. Args: coro: The coroutine function to create a task for.

Runs the code, takes a screenshot, and asks a multimodal model if it looks correct. Args: program_path: The file

Export traces to HDF5 format, which is efficient for large numerical datasets. Args: traces: A list of trace objects. output

Export traces to JSONL format (one JSON object per line). Args: traces: A list of trace objects that have a to_dict() method or are

Try to determine the programming language of a code snippet using simple heuristics. Args: code: The code snippet string. Retur

Safely calculates and formats the improvement (delta) between parent and child metrics for logging. Args: parent_metrics: A dictiona

Create a Config object from a dictionary.

Create from dictionary representation

Auto-wrap dict returns for backward compatibility

Load configuration from a YAML file.

Generate text using all available models and return a list of their responses

Generate text using a system message and conversational context

Get size of a specific artifact in bytes

Computes the text embedding for a code string or a list of code strings. Args: code (Union[str, List[str]]): The code as

Get statistics for each island

Get current tracing statistics

Increment generation counter for an island

Load traces and metadata from an HDF5 file. Args: input_path: The path to the HDF5 file. Returns: A tuple containing (l

Load traces and metadata from a single JSON file. Args: input_path: The path to the JSON file. Returns: A tuple contain

Load traces from a JSONL file. Args: input_path: The path to the JSONL file. compress: If True, the file is assumed to be gz

Log an evolution trace entry Args: iteration: Current iteration number parent_program: Parent program object

Main asynchronous entry point for the command-line interface. Returns: An integer exit code (0 for success, 1 for failure).

Perform migration between islands This should be called periodically to share good solutions between islands

Parse command-line arguments

Rebuild the models list, typically after CLI overrides change primary/secondary model fields.

Request graceful shutdown

Run evolution with process-based parallelism

Sample a program and inspirations from a specific island without modifying current_island This method is thread-safe and doesn't mod

Check if migration should occur based on generation counters

Start the process pool

Stop the process pool

Store artifacts for a program Args: program_id: ID of the program artifacts: Dictionary of artifact name to

Context manager entry

Context manager exit

Initializes the task pool with a concurrency limit.

Initializes the LLM ensemble from a list of model configurations. Args: models_cfg: A list of LLMModelConfig objects, ea

Initialize the EmbeddingClient. Args: model_name (str): The embedding model name to use (e.g., 'text-embedding-3-small')

Initialize the evolution tracer Args: output_path: Path to save trace data format: Output format ('jsonl', '

Initializes the TemplateManager, loading default and custom templates.

Initializes the prompt sampler with configuration.

Post-initialization to resolve ${VAR} env var references in api_key.

Post-initialization to set up model configurations and handle backward compatibility.

Provide a concise string representation of the result.

JSON deserializer for artifacts that handles bytes

JSON serializer for artifacts that handles bytes

Safely format improvement metrics

Safely format metrics, handling both numeric and string values

Invalidate the diversity cache when programs change significantly

Log iteration progress Args: iteration: Iteration number parent: Parent program child: Child pro

Run a single iteration in a worker process

Safely calculate average of only numeric metric values

Safely sum only numeric metric values, ignoring strings and other types

Captures the screen and returns it as a base64 encoded string. Uses pyautogui for cross-platform support.

Initialize worker process with necessary components

Add or update a fragment at runtime.

Add or update a template at runtime.

Calculate the Levenshtein edit distance between two code snippets. This measures the number of single-character edits (insertions, deletions,

Cancel all currently tracked tasks in the pool.

Evaluate multiple programs in parallel Args: programs: List of (program_code, program_id) tuples Returns:

Evolve an algorithm implemented as a class based on a benchmark function. Args: algorithm_class: The initial class to evolve.

A general-purpose function to evolve an arbitrary code string with a custom evaluator. Args: initial_code: The initial code string t

Evolve a specific Python function based on a set of test cases. Args: func: The initial function to evolve. test_cases: A li

Extract evolution traces from an existing checkpoint directory Args: checkpoint_dir: Path to checkpoint directory output_pat

Extract complete evolution traces with full lineage chains and prompts/actions This function builds the complete evolution history for each

Run a collection of tasks with a limit on how many can run concurrently. Args: n: The maximum number of tasks to run at the same tim

Generate text from a single prompt string. This is typically for completion-style models. Args: prompt (str): Th

Generate text from a single prompt string

Generate multiple texts in parallel from the same prompt

Generate text using a system message and a list of conversational messages. This is for chat-based models that support roles (system,

Get list of artifact keys

Get total size of all artifacts in bytes