Struct LinearGradientBatchArgs

pub struct LinearGradientBatchArgs<'a, T, const NI: usize, const NO: usize>
where
    T: DataTypeInfo + Debug + Default + UnitValue<T>,
{
    pub output: CudaTensor2dPtr<T, NI, NO>,
    /* private fields */
}

Defines the list that is passed to the cuda kernel function as arguments for the calculation of the amount of update of the linear layer weights during the mini-batch.

Fields

output: CudaTensor2dPtr<T, NI, NO>

Implementations

impl<'a, T, const NI: usize, const NO: usize> LinearGradientBatchArgs<'a, T, NI, NO>

where T: DataTypeInfo + Debug + Default + UnitValue<T>,

Create an instance of an object representing the list of arguments s for calculating the amount of updates to the linear layer weights during a mini-batch run.

pub fn new( loss: &'a CudaVecView<'a, T, CudaTensor1dPtr<T, NO>>, input: &'a CudaVecView<'a, T, CudaTensor1dPtr<T, NI>>, output: CudaTensor2dPtr<T, NI, NO>, batch_size: usize, ) -> LinearGradientBatchArgs<'a, T, NI, NO>

Create a LinearGradientBatchArgs instance

Arguments

• loss - loss
• input - input
• output - output (All elements must be initialized to zero.)
• batch_len - batch_count

Trait Implementations

impl<'a, T, const NI: usize, const NO: usize> KernelArgs for LinearGradientBatchArgs<'a, T, NI, NO>

where T: DataTypeInfo + Debug + Default + UnitValue<T>,

fn as_vec(&mut self) -> Vec<&mut dyn AsKernelPtr>

Returns a Vec<&mut dyn AsMutKernelPtr> of the type implementing AsMutKernelPtr, which is converted to a data type that can be passed to the cuda kernel in subsequent processing.