Expand description
Foundation crate of teenygrad: tensor/graph types, the computational
graph (graph::Graph/graph::Op/graph::Shape), neural network layers (nn), the
dtype system (dtype), device abstraction (device), and name-scoping used by every
other teeny-* crate.
no_std by default (the std feature is not in the default feature set) — enable std if
you need it. See the crate README for the full feature list.
Modules§
- compiler
- Compiler-facing traits (targets, compiled kernels).
- device
- Device abstraction (buffers, contexts, launch configuration).
- dtype
- The dtype system:
Dtype/Float/tensor traits kernel code is generic over. - errors
- Error types.
- graph
- The computational graph:
Graph,Op,Shape,SymTensor. - macros
- Internal macro helpers.
- model
- Model execution traits (
Layer, lowering, kernel-launch argument packing). - nn
- Standard neural network layers.
- runtime
- Runtime execution traits.
- utils
- Miscellaneous utilities (e.g. the
Dagtype).
Macros§
- sequential
- Chains a sequence of
Layers into a single closure, calling each in order and feeding each output into the next.