coreai_opt.quantization.ExecutionMode

class coreai_opt.quantization.ExecutionMode(value, *args, **kwargs)[source]

Bases: StrEnum

Enum representing quantization execution modes.

Each member is a string value representing the execution mode used for quantization.

Parameters:

  • value (object)

  • args (Any)

  • kwargs (Any)

Return type:

Any

GRAPH

Graph-based quantization using torch.export to capture the model as an FX graph, then applying quantization on top. Built on torchao’s PT2E implementation. Requires the model to be exportable via torch.export.export. Recommended default.

EAGER

Eager-mode quantization that works directly on nn.Module without graph capture. Supports dynamic control flow (if/else, loops) and is the fallback when a model is not exportable.

__init__(*args, **kwds)
PT2E: ExecutionMode = 'graph'

Deprecated. Use ExecutionMode.GRAPH instead.