Get the logger.

Pulls a node def from a dictionary for a given name. Args: node_map: Dictionary containing an entry indexed by name for every node.

Pulls a node def from a dictionary for a given name. Args: node_map: Dictionary containing an entry indexed by name for every node.

Move the input data to the specified device. Args: input (dict, list, tuple, or torch.Tensor): The input data to be moved.

Execute the quantize process on the specified model. Args: quant_config(dict): The user defined 'StaticQuantConfig' class.

Output log with the error level.

Construct config from a dictionary. Args: config_dict: The dictionary containing the config. Returns: The co

Get full layer name. Args: sub_layer_name (str): sub layer name block_idx (int): index of block Returns:

Return torch version if ipex exists.

Get the amount of CPU memory used by the current process in MiB (Mebibytes).

Output log with the level as a parameter.

Insert dummy biasadd for fusion.

Insert one node into the graph. Args: new_node (nodedef): new nodedef object old_output_node_names (string list):the

Validate nodes exist in the graph_def. Args: graph_def (tf.compat.v1.GraphDef): tf.compat.v1.GraphDef object. node_names (list of

Check if quantized layer method API matches original layer method API. Args: orig_cls (type): Original layer class. quant_cls (ty

Check whether version1 is less than version2.

Init RTN weight-only quantization config. Args: dtype (str): Data type for weights. Default is "int". bits (int): Num

Execute according to mode. Args: model (torch.nn.Module): The model to be executed. mode (Mode): The mode of curr

Pulls a node def from a dictionary for a given name. Args: node_map: Dictionary containing an entry indexed by name for every node.

Set module with a given op name. Args: model (object): the input model. op_name (str): name of op. new_module (object): t

Add quantized prologue nodes.

Reset output node maps.

Custom evaluate function to estimate the accuracy of the model. Args: model (tf.Graph_def): The input model graph Returns:

Load config from a JSON file. Args: filename (str): The path to the JSON file. Returns: The loaded config.

Generate the default Static quantization config. Returns: StaticQuantConfig: The default JAX Static quantization config.

Get eightbit op wise sequences information. Returns: [dictionary]: key is the op type while value is the list of sequences start

Do nothing.

Extracts the values from a const NodeDef as a numpy ndarray. Args: node_def: Const NodeDef that has the values we want to access.

Generate and save calibration info.

Get the default configuration of GPTQ. Args: processor_type (Optional[Union[str, torch_utils.ProcessorType]], optional): The user-specifi

Get op types per precision. Args: precision (string): precision name Returns: [string list]: A list composed

Get the quantizer. Initialize a quantizer or get `quantizer` attribute from model. Args: model (torch.nn.Module): pytorch model.

Convert the QTensor object to half precision.

Prunes out nodes that aren't needed for inference. Args: input_graph: Model to analyze and prune. types_to_splice: An

Save the quantized model.

Convert the configuration to a mapping. Args: config_list (List[BaseConfig]): List of base configurations. Default is None.

Detect the rule matched nodes collections. Returns: [List] -- [the matched rule] [String] -- [the list contains the m

Converts a string dtype to its corresponding PyTorch dtype. Args: str_dtype (str): The string representation of the dtype. Returns:

Output log with the debug level.

Get the dot key's item in nested dict. Usage example: person = {'person':{'name':{'first':'John'}}} deep_get(person, "person.name.f

Output log with the error level.

Evaluate the model using registered evaluation functions. Args: model: The fp32 model or quantized model. Returns:

Store an object in the memo.

Process `quantizer` attribute of model according to current phase. In `prepare` phase, the `quantizer` is set as an attribute of the model to

Quant and dequant tensor with group size. It's an in-place function. Args: weight: input weight bits (int, optional): bits. Defau

Save quantized model. Args: model (torch.nn.module or TorchScript model with IPEX or fx graph with pt2e, optional): Quantized model.

Persist a scale-related payload to disk in the chosen format. Args: model (torch.nn.Module): Model associated with the data. d (d

Create quantization helper functions for activations. When fixed_range is set, pre-computes scale (and zero point) so that call avoid

Get module with a given op name. Args: model (object): the input model. op_name (str): name of op. Returns: module (

Get the default configuration of RTN. Args: processor_type (Optional[Union[str, torch_utils.ProcessorType]], optional): The user-specifie

Load the module value. Args: model (torch.nn.module): torch model. param_name (str): module name. path (str): path to loa

Remove node with one input and rebuild internal graph data structure. Args: node_name (string): node name Returns:

Check whether version1 is greater than version2.

Delegate attribute access to the wrapped model. Args: name (str): Attribute name to access. Returns: Any: At

Delegate attribute updates to the wrapped model. Args: name (str): Attribute name to update. value (Any): Value to as

Iterates over all scale method configs (default, nodes, layers, layer_types) in the given dict and compares them to the given conditions. For

Generate optimized model by fusing BN layers or replacing Keras layers to custom quantized layers. Args: BN_fused_layers: The lay

Expand the config. case 1 { "global": { "weight_bits": [4, 6]} } expand to :

Apply the model to the input data on the specified device. Args: model (torch.nn.Module): The model to be applied. input (dic

Deserialize this class from a config dict.

Get registered configurations without priority. Returns: Dict[str, Dict[str, object]]: A dictionary containing registered configu

Obtain config for one layer, since GPTQ supports layer-wise config.

Prepares a given model for quantization. Args: model (torch.nn.Module): raw fp32 model or prepared model. example

Check whether version1 is equal to version2.

Output log with the warning level only once.