adaptive-evolve

What problem does it solve?

Adaptive-evolution workflows streamline multi-round optimization of neural network operators, reducing manual trial-and-error and uncovering high-performance configurations.

Core Features & Use Cases

• Coordinated multi-agent workflow: Designer, Coder, and Verifier collaborate to explore, implement, and validate operator optimizations.
• Adaptive search strategies: Dynamically adjust population and mutation rates to balance exploration and convergence.
• Use Case: When optimizing a fused MatMul+Bias+Activation path for a new backend, this skill coordinates design proposals, code generation, and verification to reach a performance target faster than manual tuning.

Quick Start

Run the adaptive-evolve workflow to initialize a designer-agent population, evolve designs with the coder-agent, and verify results with the verifier-agent.