ml-verify

What problem does it solve?

This Skill prevents costly mistakes by verifying ML code, configurations, and mathematical formulas against official documentation and best practices before expensive training jobs or deployments.

Core Features & Use Cases

• Pre-run Verification: Catches configuration errors, incorrect API usage, and mathematical inaccuracies.
• Documentation Grounding: Leverages a knowledge base (KB) or web scraping to cite official framework documentation for checks.
• Dry Run Checklist: Guides users through essential pre-training checks like model loading, data batching, and initial loss/gradient sanity checks.
• Use Case: Before starting a multi-day fine-tuning job, use this skill to ensure your LoRA configuration, learning rate, and data loading pipeline are correctly set up according to Hugging Face PEFT and Transformers documentation.

Quick Start

Use the ml-verify skill to check my QLoRA config for training Llama-3 8B.