remove the set_inductor_config argument of quantize_.

[vkuzo]

Summary:

Finishing the migration in #1715. Specifically:

• the set_inductor_config argument of quantize_ is removed
• the following workflows get a set_inductor_config argument, defaulting to True: AWQUIntXConfig, IntNWeightOnlyConfig, SmoothQuantConfig, Int8DynamicActivationInt4WeightConfig, Int4DynamicActivationInt4WeightConfig, GemliteUIntXWeightOnlyConfig, Int4WeightOnlyConfig, Int8WeightOnlyConfig, Int8DynamicActivationInt8WeightConfig, Float8WeightOnlyConfig, Float8DynamicActivationFloat8WeightConfig, Float8DynamicActivationFloat8WeightConfig, Float8StaticActivationFloat8WeightConfig, Float8StaticActivationFloat8WeightConfig, UIntXWeightOnlyConfig, FPXWeightOnlyConfig. I skipped quantized training and non-AQT workflows from this, since I expect them to not need this setting.

Before

def quantize_(
    model: torch.nn.Module,
    config: AOBaseConfig,
    filter_fn: Optional[Callable[[torch.nn.Module, str], bool]] = None,
    set_inductor_config: Optional[bool] = None,
    device: Optional[torch.types.Device] = None,
):  ...

After

def quantize_(
    model: torch.nn.Module,
    config: AOBaseConfig,
    filter_fn: Optional[Callable[[torch.nn.Module, str], bool]] = None,
    device: Optional[torch.types.Device] = None,
):  ...

# see top of PR description for list of affected configs
class Int8DynamicActivationInt4WeightConfig(AOBaseConfig):
    ...
    set_inductor_config: bool = True

Test Plan:

pytest test/quantization/test_quant_api.py -s -x -k test_workflow_e2e_numerics

Reviewers:

Subscribers:

Tasks:

Tags: