Graph Specifications

Curated umbrella import for GraphSpec. Use this import when working with GraphSpec models, primitives, lowering, and bridge theorems:

import NN.GraphSpec

It gives you:

• the canonical DAG model API (NN.GraphSpec.DAG.Term, NN.GraphSpec.DAG.Model),
• the sequential authoring syntax (NN.GraphSpec.Chain + >>>) for chain models and its lowering into DAG (Chain.ToDAG.Model), with the theorem that the converted term has the direct chain interpretation (Chain.ToDAG.Semantics),
• the Spec semantics (NN.GraphSpec.Interp.spec) and TorchLean lowering (NN.GraphSpec.Chain.toProgram),
• sequential and DAG primitive packs. The small primitives (linear, relu, softmax) live in Chain.Primitives; larger packs such as Primitives.Spatial (convolution, pooling, flattening, batch normalization) and Primitives.Embedding sit under NN/GraphSpec/Primitives/,
• the GraphSpec example architectures (NN.GraphSpec.Models),
• the optional lowering to Runtime.Autograd.Model.Layers.Seq when primitives provide toLayerM?,
• and the model/primitive bridge theorems that connect GraphSpec syntax to Spec references.

Sequential Chain pipelines can be lowered to the canonical NN.GraphSpec.DAG.Model via NN.GraphSpec.LowerToDAG.Chain.toDAGTerm and NN.GraphSpec.LowerToDAG.Chain.toDAGModelZeroInit.

Umbrella re-export; the implementation lives in the imported modules.