When should I load this skill instead of a specific optimizer skill?

Load this skill when you encounter training problems but aren't yet sure if the cause is the learning rate, the optimizer, the data, or the gradients.

How does it handle overfitting issues?

It provides a multi-pronged approach, routing you to regularization techniques, data augmentation strategies, and hyperparameter tuning tools based on your dataset size.

Does it fix implementation bugs in PyTorch or TensorFlow?

No, this skill focuses on training dynamics and mathematical optimization. For framework-specific implementation bugs, use specialized engineering skills.

Can it help with low GPU utilization?

Yes, it includes diagnostics to determine if your training is bottlenecked by data loading, CPU preprocessing, or memory constraints.

What is the primary purpose of this skill?

It acts as a diagnostic hub that identifies specific training problems (like instability or slow convergence) and routes you to the correct optimization sub-skill.

ML Training Optimization Router

Name: ML Training Optimization Router
Author: tachyon-beep

bytachyon-beep

•

Data Science & ML

Diagnoses machine learning training issues and routes users to specific optimization strategies based on model symptoms.

About

The ML Training Optimization Router serves as a specialized diagnostic hub for troubleshooting deep learning training pipelines. It analyzes common symptoms—such as flat loss curves, gradient instability (NaN values), overfitting, and low GPU throughput—to recommend the most effective optimization techniques. By providing structured diagnostic questions and systematic routing to specialized sub-skills like gradient management, learning rate scheduling, or data augmentation, it ensures developers apply the right fixes to complex convergence and performance problems without wasting time on trial and error.

Key Features

Strategies for resolving overfitting and large validation gaps
5 GitHub stars
Performance profiling to identify bottlenecks in GPU utilization
Guidance on hyperparameter selection and experiment tracking setup
Troubleshooting for numerical instability and NaN gradient values
Symptom-based diagnostic routing for complex training failures

Use Cases

Diagnosing why a model's loss is stuck or not decreasing during the initial epochs
Fixing numerical instability and gradient explosions in deep neural networks
Optimizing training throughput and memory usage to handle larger batch sizes

About

Key Features

Strategies for resolving overfitting and large validation gaps
5 GitHub stars
Performance profiling to identify bottlenecks in GPU utilization
Guidance on hyperparameter selection and experiment tracking setup
Troubleshooting for numerical instability and NaN gradient values
Symptom-based diagnostic routing for complex training failures

Use Cases

Diagnosing why a model's loss is stuck or not decreasing during the initial epochs
Fixing numerical instability and gradient explosions in deep neural networks
Optimizing training throughput and memory usage to handle larger batch sizes