Experiment Advisor

소개

Experiment Advisor is designed for researchers and machine learning engineers who need guidance on the next steps of their development cycle. By scanning local training reports, experiment logs, and troubleshooting documentation, the skill generates 2–5 concrete experiment proposals tailored to your hardware and budget constraints. It ensures continuity in research by leveraging successful defaults from existing skills while explicitly avoiding documented failure patterns to optimize the iteration process.

주요 기능

• Integrates with troubleshooting documentation to avoid known errors
• 0 GitHub stars
• Analyzes historical training reports and experiment logs for context
• Optimizes plans based on hardware constraints and run budgets
• Generates 2–5 structured experiment proposals with specific hyperparameters
• Appends summarized plans to the experiment log for team visibility

사용 사례

• Designing data mixture experiments for complex machine learning tasks
• Deciding on hyperparameter adjustments after reviewing recent training logs
• Planning model architecture iterations based on documented performance benchmarks