Extract failure artifacts from diagnostics for a given example. Args: diagnostics: Diagnostics dictionary from pass_at_2_accuracy_si

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

Format feature coordinates for display in prompts Args: metrics: All metrics from evaluation feature_dimensions: List of MAP

Generate text using a randomly selected model based on weights

Gets the outward normals of a polygon's edges.

Gets the outward normals of a polygon's edges.

List all available AlgoTune tasks.

Load the baseline MLIR implementation

Load configuration from YAML file.

Load HuggingFace dataset based on configuration.

Load the prompt from text file and dataset config from matching _dataset.yaml file.

Log a prompt for a program. Only logs if self.config.log_prompts is True. Args: program_id: ID of the program to log

Log in to Kattis using the access information in a kattisrc file Returns a requests.Response with cookies needed to be able to submit

Main entry point Returns: Exit code

Move to the next island in round-robin fashion

Define attention optimization parameters for evolution. The goal is to achieve 32% speedup (1.32x) like AlphaEvolve paper by optimiz

Extract a full rewrite from an LLM response Args: llm_response: Response from the LLM language: Programming language Re

Compute pass@2 accuracy across multiple ARC test cases. Args: all_attempts: List of lists of 2 numpy arrays for each test case.

Print a summary table of all results

Projects a polygon onto an axis and returns the min/max values.

Projects a polygon onto an axis and returns the min/max values.

Custom Metal kernel implementation for Qwen3 GQA attention. Args: queries: [B, num_heads=16, L, head_dim=128] keys: [B, num_

Custom Metal kernel implementation for Qwen3 GQA attention. Args: queries: [B, num_heads=16, L, head_dim=128] keys: [B, num_

Read program from file.

Request graceful shutdown

Run iterative refinement loop. Returns: dict with results including iterations, scores, solution_found_at

Run mlx-lm generation and parse the output

Run a quick test with just a few key benchmarks with proper warmup

Run a function with a timeout using concurrent.futures

Run the program in a separate process with timeout using a simple subprocess approach Args: program_path: Path to the program fi

Run the program in a separate process with timeout using a simple subprocess approach Args: program_path: Path to the program fi

Run a function with a timeout using concurrent.futures Args: func: Function to run args: Arguments to pass to the function

Runs a function with a specified timeout, raising an exception on timeout.

Set which island is currently being evolved

Set custom templates to use for this sampler Args: system_template: Template name for system message user_te

Solve the fft_cmplx_scipy_fftpack problem. Args: problem: Dictionary containing problem data specific to fft_cmp

Solve the eigenvectors_complex problem. Args: problem: Dictionary containing problem data specific to eigenvecto

Compute the 2D convolution of arrays a and b using "full" mode and "fill" boundary. Uses scipy.signal.fftconvolve for efficiency, whi

Solve the affine_transform_2d problem. Args: problem: Dictionary containing problem data specific to affine_tran

Solve the polynomial_real problem using a JIT-compiled JAX function. Args: problem: A list of polynomial coeffic

Solve the psd_cone_projection problem. Args: problem: Dictionary containing problem data specific to psd_cone_pr

Solve the fft_convolution problem. Args: problem: Dictionary containing problem data specific to fft_convolution

Computes the LU factorization of a matrix using an optimized scipy call.

Route diff blocks to either code or changes_description based on exact line-wise match of SEARCH text. Returns (code_blocks, changes_desc_blo

Calculate theoretical pass@k for uniform random search.

Convert to dictionary representation

Validate that circles don't overlap and are inside the unit square Args: centers: np.array of shape (n, 2) with (x, y) coordinates

Validate that circles don't overlap and are inside the unit square Args: centers: np.array of shape (n, 2) with (x, y) coordinates

Write program to file.

Analyze performance with enhanced safety metrics

Safely apply attention hook

Load an evolved hook program and apply its attention optimization. The hook program must expose a `create_metal_qwen3_optimization_hook()` f

Apply stochastic variations to the template

Build monomial-basis polynomial P from Hermite-basis coefficients.

Render messages into a single plain-text prompt for the manual UI.

Bulletproof extraction with comprehensive Metal kernel validation

Compute r_max and C4 from full Hermite coefficient vector (Google-style).

Cache a diversity value with LRU eviction

Calculate diversity within an island (deterministic version)

Calculate final score adjusted for safety

Clean up extracted __init__ method body by removing docstrings and super() calls. Keep only the actual initialization statements.

Remove artifact directories older than the configured retention period. Args: checkpoint_path: The path of the current c

Command-buffer-protected benchmarking with maximum safety

Given m input coeffs (c0..c_{m-1}), build the Hermite combo c0*H0 + ... + c_{m-1}*H_{4(m-1)} + c_last*H_{4m} where c_last is chosen so th

Adapted from SakanaAI/ShinkaEvolve (Apache-2.0 License) Original source: https://github.com/SakanaAI/ShinkaEvolve/blob/main/shinka/da

Create artifact directory for a program

Create bulletproof execution environment with enhanced imports

Create maximum length context prompt for stress testing

Create medium-length context prompt

Create prompt that builds context progressively

Distribute loaded programs across islands when no island metadata exists

Enforce the population size limit by removing worst programs if needed Args: exclude_program_id: Program ID to never rem

Ensure standard attention is active

Return an error result.

Evaluate a Rust sorting algorithm implementation. Tests the algorithm on various data patterns to measure: - Correctness - Performan

Return an exception result.

Extract the expected problem keys from the solve method.

Format the evolution history for the prompt

Format metrics for the prompt using safe formatting

Generate comprehensive evaluation summary with safety info