Expand description
Lazy computation graph IR.
Tensors are handles to nodes in this graph. Computation is deferred until
eval() is called, at which point the scheduler topologically sorts the
graph and dispatches to the active backend.
Structs§
- Graph
- The computation graph arena.
- Node
- A node in the lazy computation graph.
- NodeId
- Unique identifier for a node in the computation graph.
- Tensor
Meta - Metadata about a tensor (known before materialization).
Enums§
- OpKind
- The set of operations supported by the graph IR.