ExTorch.Export.Model (extorch v0.4.0)

Copy Markdown

A loaded ExportedProgram model ready for inference.

Contains the computation graph, weight tensors, and input/output mappings.

The compiled_graph field holds per-node closures built at load/1 time that capture pre-resolved scalar args and tensor input refs, so forward/2 is a tight loop over closures instead of re-parsing the graph each call. initial_values holds the pre-built parameter value map so forward doesn't re-resolve p*/b* names on every inference.

Summary

Types

t()

Types

t()

@type t() :: %ExTorch.Export.Model{
  compiled_graph: [{[String.t()], (map() -> any())}],
  device: atom() | {atom(), non_neg_integer()},
  initial_values: map(),
  native_compiled: ExTorch.Export.CompiledGraph.t() | nil,
  param_inputs: [String.t()],
  schema: map(),
  user_inputs: [String.t()],
  weights: %{required(String.t()) => ExTorch.Tensor.t()}
}