How does it help with GPU memory issues?

The skill identifies frequent CPU-GPU transfers, missing gradient accumulation for large batches, and suggests the use of mixed precision to prevent Out-of-Memory (OOM) errors.

Can it detect data leakage in my preprocessing steps?

Yes, it specifically looks for patterns like calling fit_transform on test data or improper validation splits that lead to over-optimistic results.

Which ML frameworks does this skill support?

The ML Reviewer skill supports PyTorch, TensorFlow, Keras, and scikit-learn, including MLOps tools like Hydra, Weights & Biases, and MLflow.

Does this skill provide code examples for fixes?

Yes, when it identifies an issue like missing gradient clipping or improper DataLoader settings, it provides 'Good' vs 'Bad' code comparisons for easy implementation.

ML Code Reviewer

Name: ML Code Reviewer
Author: physics91

byphysics91

0•

Data Science & ML

Reviews machine learning and deep learning code to optimize model architecture, training loops, and data pipelines for PyTorch and TensorFlow.

The ML Code Reviewer is a specialized skill for Claude Code designed to audit and improve machine learning projects. It provides domain-specific guidance for popular frameworks like PyTorch, TensorFlow, and scikit-learn, identifying critical issues such as data leakage, missing inference modes, and GPU bottlenecks. Whether you are transitioning a research script to production or troubleshooting training performance, this skill ensures your ML code follows industry best practices for efficiency, reproducibility, and MLOps standards.

Key Features

01Validation of MLOps patterns including experiment tracking and seed setting

02Detection of critical ML bugs like data leakage and missing model.eval() calls

030 GitHub stars

04Framework-specific auditing for PyTorch, TensorFlow, Keras, and scikit-learn

05Automated data pipeline checks for OOM risks and preprocessing efficiency

06GPU performance optimization and memory leak identification

Use Cases

01Reviewing deep learning training scripts for performance and correctness before long-running jobs

02Debugging non-reproducible experiments and inconsistent model inference results

03Optimizing data loading pipelines to resolve CPU bottlenecks and maximize GPU utilization

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add physics91/claude-vibe ml-reviewer

For use in Claude.ai and ChatGPT

Download Skill