Expand description
§A WebGPU based Tensor Library
This library is intended as a pure Rust Tensor
library which runs on the GPU everywhere.
In order to achieve it, it build upon the wgpu
crate which can target any platform supporting
either Vulkan
, Metal
or DX12
. It can run even on a Raspberry Pi 4 GPU.
The API entry point is the Tensor
structure.
Re-exports§
pub use crate::GpuAllocated;
Modules§
Macros§
Structs§
- CpuTensor
- GpuStore
- GpuTensor
- GpuTensor
View - A GpuTensorView share the same data as the original Tensor, but can have different shapes and strides For example, the original shape could be [2, 2] and the GpuTensorView could be [1, 2, 2]
- GpuTensor
View Mut - Same as GpuTensorView but mutable
- Linear
Indexer - Shape
Strides - Slice
Range Info - Tensor
- A Tensor is an N dimensional data structure. This is the entry point for most of the API
of this crate. This is normally backed by GPU memory and its device chosen using the current
default of the
crate::GpuStore::get_default()
, which can be changed however, one can NOT do operations using two Tensors from different devices. - Tensor
View - A view into the original
Tensor
. A view borrows part of (or even the entirety of) the original Tensor data. It can NOT modify the data itself. It is normally produced by indexing or slicing a Tensor.