Expand description
NumRs — core library
This crate provides the core runtime, IR, lowering pipeline and basic backends to run simple numeric kernels. It’s intentionally small and focused on proving the architecture end-to-end.
Public surface will be stable: array creation + add/mul/sum
Re-exports§
pub use array::cast_array;pub use array::promoted_dtype;pub use array::Array;pub use array::DType;pub use array::DTypeValue;pub use array_view::ArrayView;pub use autograd::is_grad_enabled;pub use autograd::set_grad_enabled;pub use autograd::NoGrad;pub use autograd::Tensor;pub use autograd::AdaGrad;pub use autograd::Adam;pub use autograd::Optimizer;pub use autograd::RMSprop;pub use autograd::SGD;pub use autograd::BatchNorm1d;pub use autograd::Conv1d;pub use autograd::Dropout;pub use autograd::Flatten;pub use autograd::Linear;pub use autograd::Module;pub use autograd::ReLU;pub use autograd::Sequential;pub use autograd::Sigmoid;pub use autograd::CrossEntropyLoss;pub use autograd::Dataset;pub use autograd::MSELoss;pub use autograd::Trainer;pub use autograd::TrainerBuilder;pub use backend::dispatch::get_backend_override;pub use backend::dispatch::set_backend_override;pub use llo::reduction::ReductionKind;pub use llo::ElementwiseKind;pub use ops::abs;pub use ops::abs;pub use ops::acos;pub use ops::acos;pub use ops::add;pub use ops::asin;pub use ops::asin;pub use ops::atan;pub use ops::atan;pub use ops::cos;pub use ops::cos;pub use ops::div;pub use ops::exp;pub use ops::exp;pub use ops::log;pub use ops::log;pub use ops::mul;pub use ops::pow;pub use ops::relu;pub use ops::relu;pub use ops::sigmoid;pub use ops::sigmoid;pub use ops::sin;pub use ops::sin;pub use ops::softmax;pub use ops::softmax;pub use ops::sqrt;pub use ops::sqrt;pub use ops::sub;pub use ops::sum;pub use ops::sum;pub use ops::tan;pub use ops::tan;pub use ops::tanh;pub use ops::tanh;pub use startup::print_startup_log;
Modules§
- array
- Array module public API
- array_
view - ArrayView - Zero-copy view into typed data for FFI operations
- autograd
- Automatic differentiation system for NumRs
- backend
- Backend selection and execution helpers
- codegen
- ir
- llo
- Low-level operations (LLO) — target-aware operations ready for codegen
- ops
- NumRs Operations API
- ops_
inplace - Zero-copy in-place operations for FFI bindings
- startup
Macros§
- no_grad
- Macro para ejecutar código sin gradientes