Does it support specific frameworks like PyTorch or TensorFlow?

The diagnostic logic is framework-agnostic, and the provided code examples are designed for standard Python-based machine learning environments.

How does it analyze my model's gradients?

The skill provides Python snippets to inspect per-layer gradient norms, helping you identify which parts of the architecture are causing numerical instability.

Can it help with learning rate optimization?

Yes, it analyzes the correlation between your learning rate schedule and loss behavior to suggest if your LR is too high, too low, or needs a warmup phase.

What training issues can this skill help diagnose?

It is designed to handle loss plateaus, NaN/Inf errors, gradient vanishing or explosion, overfitting, and unstable convergence.

AI Training Diagnostics

Name: AI Training Diagnostics
Author: rHedBull

byrHedBull

数据科学与机器学习

Diagnoses and fixes machine learning training issues including loss plateaus, gradient instabilities, and convergence failures.

关于

AI Training Diagnostics provides a systematic, interactive workflow for troubleshooting neural network training when performance stalls or fails. It guides users through symptom identification—such as NaN losses, exploding gradients, or overfitting—and executes Python-based diagnostic checks to analyze gradient norms and loss curves. By generating ranked hypotheses and offering automated validation of fixes, this skill helps developers move from training failure to convergence with high confidence and minimal trial-and-error.

主要功能

Automated gradient norm analysis and layer-wise health checks
0 GitHub stars
Interactive symptom-based diagnostic workflow for ML training
Visual loss curve and learning rate correlation plotting
Ranked hypothesis generation for identifying root causes
Post-fix validation loop to verify training improvements

使用场景

Debugging gradient vanishing or explosion in deep architectures
Troubleshooting NaN or Inf loss values during model initialization
Identifying the cause of training plateaus or oscillating loss curves

关于

主要功能

Automated gradient norm analysis and layer-wise health checks
0 GitHub stars
Interactive symptom-based diagnostic workflow for ML training
Visual loss curve and learning rate correlation plotting
Ranked hypothesis generation for identifying root causes
Post-fix validation loop to verify training improvements

使用场景

Debugging gradient vanishing or explosion in deep architectures
Troubleshooting NaN or Inf loss values during model initialization
Identifying the cause of training plateaus or oscillating loss curves