w4¶
- w4(*, axis=None, execution_mode=<ExecutionMode.GRAPH: 'graph'>)
int4 weight-only quantization, per-channel symmetric.
- Parameters:
axis (int | None) – Channel axis for per-channel quantization. When
None(default), the axis is auto-resolved based on the module type during quantization.execution_mode (ExecutionMode) – Quantization execution mode. Defaults to
ExecutionMode.GRAPH.
- Returns:
int4 weight-only configuration.
- Return type: