Crate shrew_core 

shrew-core

Core tensor primitives, types, backend traits, and autograd for Shrew.

This crate provides:

• Tensor — n-dimensional array with automatic differentiation
• Shape / Layout — shape, strides, and memory layout
• DType — data types (F16, BF16, F32, F64, U8, U32, I64)
• Backend trait — abstraction over CPU/GPU execution
• GradStore — gradient storage returned by backward()

Re-exports

pub use backend::Backend;

pub use backend::BackendDevice;

pub use backprop::GradStore;

pub use backprop::checkpoint;

pub use backprop::checkpoint_sequential;

pub use backprop::is_checkpointing;

pub use backprop::with_checkpoint_mode;

pub use dtype::DType;

pub use dtype::WithDType;

pub use dynamic_shape::ShapeEnv;

pub use dynamic_shape::ShapeGuard;

pub use dynamic_shape::SymDim;

pub use dynamic_shape::SymbolicShape;

pub use error::Error;

pub use error::Result;

pub use layout::Layout;

pub use op::Op;

pub use shape::Shape;

pub use tensor::Tensor;

Modules

backend

backprop

dtype

dynamic_shape

error

layout

op

shape

tensor

Macros

bail

Macro for early return with a formatted error message. Usage: bail!("something went wrong: {}", detail)