croq-tune-iterative

What problem does it solve?

This Skill provides a durable wrapper for GPU kernel tuning, enabling checkpointed workflows with human confirmation at invocation boundaries to improve process reliability and control.

Core Features & Use Cases

• Checkpointing: Saves the tuning session state before and after each autonomous tuning run to ensure recoverability.
• Human-in-the-loop: Allows user intervention and decision-making between tuning invocations without disrupting the underlying tuning process.
• Use Case: A researcher tuning GPU kernels for different configurations can pause, inspect, and resume tuning sessions seamlessly, ensuring progress is preserved even in case of interruptions.

Quick Start

Invoke the wrapper with desired parameters, then follow prompts for checkpoints and continuation of the tuning process.