Namespace AiDotNet.MixedPrecision

Classes

LossScaler<T>

Implements dynamic loss scaling for mixed-precision training to prevent gradient underflow.

MixedPrecisionConfig

Configuration settings for mixed-precision training.

MixedPrecisionContext

Manages master weights (FP32) and working weights (FP16) for mixed-precision training.

MixedPrecisionTrainingLoop<T>

Implements mixed-precision training loop for neural networks following NVIDIA's approach.