Shortcuts

QConfigMapping

class torch.ao.quantization.qconfig_mapping.QConfigMapping[source]

Mapping from model ops to torch.ao.quantization.QConfig s.

The user can specify QConfigs using the following methods (in increasing match priority):

set_global : sets the global (default) QConfig

set_object_type : sets the QConfig for a given module type, function, or method name

set_module_name_regex : sets the QConfig for modules matching the given regex string

set_module_name : sets the QConfig for modules matching the given module name

set_module_name_object_type_order : sets the QConfig for modules matching a combination of the given module name, object type, and the index at which the module appears

Example usage:

qconfig_mapping = QConfigMapping()
    .set_global(global_qconfig)
    .set_object_type(torch.nn.Linear, qconfig1)
    .set_object_type(torch.nn.ReLU, qconfig1)
    .set_module_name_regex("foo.*bar.*conv[0-9]+", qconfig1)
    .set_module_name_regex("foo.*", qconfig2)
    .set_module_name("module1", qconfig1)
    .set_module_name("module2", qconfig2)
    .set_module_name_object_type_order("foo.bar", torch.nn.functional.linear, 0, qconfig3)
classmethod from_dict(qconfig_dict)[source]

Create a QConfigMapping from a dictionary with the following keys (all optional):

“” (for global QConfig)

“object_type”

“module_name_regex”

“module_name”

“module_name_object_type_order”

The values of this dictionary are expected to be lists of tuples.

Return type

QConfigMapping

set_global(global_qconfig)[source]

Set the global (default) QConfig.

Return type

QConfigMapping

set_module_name(module_name, qconfig)[source]

Set the QConfig for modules matching the given module name. If the QConfig for an existing module name was already set, the new QConfig will override the old one.

Return type

QConfigMapping

set_module_name_object_type_order(module_name, object_type, index, qconfig)[source]

Set the QConfig for modules matching a combination of the given module name, object type, and the index at which the module appears.

If the QConfig for an existing (module name, object type, index) was already set, the new QConfig will override the old one.

Return type

QConfigMapping

set_module_name_regex(module_name_regex, qconfig)[source]

Set the QConfig for modules matching the given regex string.

Regexes will be matched in the order in which they are registered through this method. Thus, the caller should register more specific patterns first, e.g.:

qconfig_mapping = QConfigMapping()
    .set_module_name_regex("foo.*bar.*conv[0-9]+", qconfig1)
    .set_module_name_regex("foo.*bar.*", qconfig2)
    .set_module_name_regex("foo.*", qconfig3)

In this example, “foo.bar.conv0” would match qconfig1, “foo.bar.linear” would match qconfig2, and “foo.baz.relu” would match qconfig3.

If the QConfig for an existing module name regex was already set, the new QConfig will override the old one while preserving the order in which the regexes were originally registered.

Return type

QConfigMapping

set_object_type(object_type, qconfig)[source]

Set the QConfig for a given module type, function, or method name. If the QConfig for an existing object type was already set, the new QConfig will override the old one.

Return type

QConfigMapping

to_dict()[source]

Convert this QConfigMapping to a dictionary with the following keys:

“” (for global QConfig)

“object_type”

“module_name_regex”

“module_name”

“module_name_object_type_order”

The values of this dictionary are lists of tuples.

Return type

Dict[str, Any]

Docs

Access comprehensive developer documentation for PyTorch

View Docs

Tutorials

Get in-depth tutorials for beginners and advanced developers

View Tutorials

Resources

Find development resources and get your questions answered

View Resources