debug:tensorflow

To fix TensorFlow GPU memory issues, you need to diagnose device availability, enable memory growth, and verify CUDA compatibility. This skill systematically detects GPU/CUDA problems and optimizes memory management for training pipelines.

TensorFlow training loss returns NaN or Inf due to numerical instability in loss calculations or metrics. This skill identifies the source of NaN/Inf losses and provides mitigation strategies to stabilize model training.

Resolving tensor shape and dtype mismatches in Keras requires validating that tensor shapes and data types align with model expectations. This skill quickly verifies tensor configurations to prevent structural errors during model development.

Identifying bottlenecks in tf.data pipelines involves profiling data throughput to isolate slow operations. This skill analyzes your data pipeline to pinpoint performance constraints and optimize end-to-end training throughput.

You can diagnose TensorFlow gradient issues by inspecting gradient flow and applying clipping to address training stalls. This skill analyzes gradient behavior to help resolve stalled training sessions and improve convergence.

Fixing SavedModel loading errors requires ensuring correct serialization formats and proper handling of custom objects. This skill verifies compatibility to resolve deployment issues when loading TensorFlow models.