MLCompute iOS 14.5 to 15.0 API Differences

MLCAdamOptimizer.h

Added MLCAdamOptimizer.usesAMSGrad

Added +[MLCAdamOptimizer optimizerWithDescriptor:beta1:beta2:epsilon:usesAMSGrad:timeStep:]

MLCAdamWOptimizer.h

Added MLCAdamWOptimizer

Added MLCAdamWOptimizer.beta1

Added MLCAdamWOptimizer.beta2

Added MLCAdamWOptimizer.epsilon

Added MLCAdamWOptimizer.usesAMSGrad

Added MLCAdamWOptimizer.timeStep

Added +[MLCAdamWOptimizer optimizerWithDescriptor:]

Added +[MLCAdamWOptimizer optimizerWithDescriptor:beta1:beta2:epsilon:usesAMSGrad:timeStep:]

MLCDevice.h

Added MLCDevice.actualDeviceType

Added +[MLCDevice aneDevice]

MLCLayer.h

Added MLCLayer.deviceType

MLCOptimizer.h

Added MLCOptimizer.gradientClippingType

Added MLCOptimizer.maximumClippingNorm

Added MLCOptimizer.customGlobalNorm

MLCOptimizerDescriptor.h

Added MLCOptimizerDescriptor.gradientClippingType

Added MLCOptimizerDescriptor.maximumClippingNorm

Added MLCOptimizerDescriptor.customGlobalNorm

Added +[MLCOptimizerDescriptor descriptorWithLearningRate:gradientRescale:appliesGradientClipping:gradientClippingType:gradientClipMax:gradientClipMin:maximumClippingNorm:customGlobalNorm:regularizationType:regularizationScale:]

MLCPlatform.h

Added MLCPlatform

Added +[MLCPlatform setRNGSeedTo:]

Added +[MLCPlatform getRNGseed]

MLCTensor.h

Added +[MLCTensor tensorWithShape:randomInitializerType:dataType:]

Added -[MLCTensor tensorByQuantizingToType:scale:bias:]

Added -[MLCTensor tensorByQuantizingToType:scale:bias:axis:]

Added -[MLCTensor tensorByDequantizingToType:scale:bias:]

Added -[MLCTensor tensorByDequantizingToType:scale:bias:axis:]

MLCTypes.h

Added MLCDataTypeFloat16

Added MLCDataTypeInt8

Added MLCDataTypeUInt8

Added MLCDeviceTypeANE

Added MLCExecutionOptionsPerLayerProfiling

Added MLCGradientClippingType

Added MLCGradientClippingTypeByValue

Added MLCGradientClippingTypeByNorm

Added MLCGradientClippingTypeByGlobalNorm

Added MLCGradientClippingTypeDebugDescription()