Can it fix NaN or Inf values in my training loss?

Yes, it offers diagnostic steps to enable numeric checking, implement gradient clipping, and use numerically stable operations to prevent training instability.

Does it support GPU-specific issues?

Absolutely. It includes checks for CUDA/cuDNN compatibility, NVIDIA driver verification, and scripts to force GPU visibility in multi-GPU environments.

How does this skill help with TensorFlow OOM errors?

It provides specific code patterns to enable memory growth, limit GPU memory allocation, and implement mixed precision training to reduce the memory footprint of your models.

How does it handle shape mismatches?

It uses tf.debugging assertions and eager execution strategies to pinpoint where dimensions diverge and suggests precise reshaping or expansion fixes.

TensorFlow Debugger

Name: TensorFlow Debugger
Author: SnakeO

bySnakeO

데이터 과학 및 ML

Diagnoses and resolves complex machine learning issues in TensorFlow and Keras workflows systematically.

소개

This skill provides a comprehensive framework for debugging TensorFlow applications, helping developers identify and fix common machine learning pitfalls such as tensor shape mismatches, GPU configuration errors, and numerical instabilities like NaN or Inf values. It leverages advanced tools including tf.debugging assertions, TensorBoard profiling, and memory management strategies to optimize data pipelines and model training. Whether you are facing Out-of-Memory (OOM) errors, vanishing gradients, or SavedModel version conflicts, this skill guides you through reproduction, isolation, and resolution phases to ensure robust model performance.

주요 기능

GPU and CUDA detection failure troubleshooting
Automated diagnosis of tensor shape and dimension mismatches
TensorBoard integration and performance profiling
0 GitHub stars
Memory optimization for Out-of-Memory (OOM) errors
Numerical stability checking for NaN/Inf loss values

사용 사례

Fixing training crashes caused by GPU memory exhaustion
Resolving SavedModel loading and version compatibility issues
Debugging vanishing or exploding gradients in deep neural networks

소개

주요 기능

GPU and CUDA detection failure troubleshooting
Automated diagnosis of tensor shape and dimension mismatches
TensorBoard integration and performance profiling
0 GitHub stars
Memory optimization for Out-of-Memory (OOM) errors
Numerical stability checking for NaN/Inf loss values

사용 사례

Fixing training crashes caused by GPU memory exhaustion
Resolving SavedModel loading and version compatibility issues
Debugging vanishing or exploding gradients in deep neural networks