Struct CudaTensor 

pub struct CudaTensor { /* private fields */ }

GPU-resident tensor with gradient support

This is the CUDA-accelerated replacement for Tensor when the cuda feature is enabled.

Implementations

impl CudaTensor

pub fn from_vec( device: &CudaDevice, data: Vec<f32>, requires_grad: bool, ) -> Result<Self>

Create a new tensor on GPU from host data

pub fn zeros( device: &CudaDevice, len: usize, requires_grad: bool, ) -> Result<Self>

Create a tensor filled with zeros

pub fn ones( device: &CudaDevice, len: usize, requires_grad: bool, ) -> Result<Self>

Create a tensor filled with ones

pub fn to_vec(&self) -> Result<Vec<f32>>

Copy tensor data back to CPU

pub fn grad_to_vec(&self) -> Result<Option<Vec<f32>>>

Get gradient as CPU vector (if computed)

pub fn copy_from_vec(&mut self, data: &[f32]) -> Result<()>

Update data from CPU vector

pub fn set_grad_from_vec(&mut self, grad: &[f32]) -> Result<()>

Set gradient from CPU vector

pub fn zero_grad(&mut self) -> Result<()>

Zero out gradient

pub fn requires_grad(&self) -> bool

Check if requires gradient

pub fn len(&self) -> usize

Get number of elements

pub fn is_empty(&self) -> bool

Check if empty

pub fn data_buffer(&self) -> &GpuBuffer<f32>

Get raw GPU buffer for data (for kernel operations)

pub fn data_buffer_mut(&mut self) -> &mut GpuBuffer<f32>

Get mutable raw GPU buffer for data

pub fn grad_buffer(&self) -> Option<&GpuBuffer<f32>>

Get raw GPU buffer for gradient (for kernel operations)

pub fn grad_buffer_mut(&mut self) -> Option<&mut GpuBuffer<f32>>

Get mutable raw GPU buffer for gradient

pub fn device(&self) -> &Arc<CudaContext>

Get device context

Trait Implementations

impl Debug for CudaTensor

Available on crate feature cuda only.

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.