r"""Prepares a copy of the model for quantization calibration or quantization-aware training and converts it to quantized version. Quantizati

Prints only on the master process.

Quantize tensor to NF4/FP4 data type. Args: tensor: input tensor quantile (float, optional): percentile of clip. Defaults to 1.0.

A function used for element-wise quantization with mx_specs. Args: A (torch.Tensor): a tensor that needs to be quantized mx_specs

Register supported configs for specific framework. Args: fwk_name: the framework name.

Remove sub-graph with a const node. Args: new_node (nodedef): the constant node old_end_node_name (string): the sub-

This is neural_compressor function include tuning, export and benchmark option.

This is neural_compressor function include tuning, export and benchmark option.

This is neural_compressor function include tuning, export and benchmark option.

This is neural_compressor function include tuning, export and benchmark option.

This is neural_compressor function include tuning and benchmark option

Save task to a preset directory. Args: preset_dir: The path to the local model preset directory. max_shard_size: `int

Set new module into model by key name. Args: model (torch.nn.Module): original model key (str): module name to be replaced

The function is used for ipex warm-up inference.

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

Change the option for qdq.

Check if the given object is the given supported type and in the given supported value. Example:: from neural_compressor.common.base_con

Get the loss for auto-tuning. Args: output (Tensor): FP32 output for one layer. output_q (Tensor): Quantized outp

Obtain the optimal alpha values based on shared criteria and loss values recorded in the auto-tuning step. Returns: dict: A di

The version 1 function that get graph from saved_model. Args: model (string or tf.keras.Model): model path or tf.keras.Model object.

Get nodes from graph_def using node name. Args: graph_def (graph_def): graph_def node_name (str): node name Returns:

Wrapper for inserting int64 features into Example proto.

Load weight file with safe defaults. `weights_only=True` avoids unpickling arbitrary Python objects when loading quantized checkpoint

Pad x and y so that the results have the same length (second dimension).

Check whether hpu weight-only quantization linear module can be used. return True when: 1. device is 'hpu' 2. model has hpu f

Get Reshape input nodes.

Clean module weight.

Converts a sequence of [tokens|ids] using the vocab.

Create a nodedef object. Args: op (string): op type name (string): op name inputs (string list): op's inp

Create a simple model for testing.

Create test data for inference.

Converts a dictionary specifying scale method parameters (granularity, scale_value_type, rounding_method, backoff, params) into a ScaleMethod

Abstract method of model conversion to ONNX.

Get all layers with target types.

Find scale and zero for weight.

Fix ref type of the graph_def.

Create a new HQQModule instance from a floating-point linear. Args: float_module (torch.nn.Linear): The floating-point module to

Generate feed dict helper function.

Static method that generate int32 bias for matmul op. Args: bias_tensor: bias node tensor. weights_tensor: weights te

Get accelerator by name.

Get the default configuration set for tuning. Returns: Union[None, "RTNConfig", List["RTNConfig"]]: The configuration set for tun

Get the frame info of the model. Returns: [parent_frame_details]: OrderedDict frame info of the graph nodes.

Get model info according to white_module_list.

Check the specified node has positive input data or not. Args: node_name (string): node name Returns: bool:

Return whether ipex is available.

Check the model_path format is saved_model or not. Args: model_path (string): the model folder path Returns: bool: return Tr

Apply operations to model layers matching the filter function. Args: model (keras.Model): Keras model with a _flatten_layers iterator.

Load a empty model.

Collect output of module. Args: model (torch.nn.Module): input model. inputs (list): a list of module input.

Normalize the answer text. Lower text, remove punctuation, articles and extra whitespace, and replace other whitespace (newline, tab, etc.) t

Parse a input saved_model. Args: model(string or AutoTrackable object): The input saved_model. Returns: graph_def: The graph

Prepare this model with the given (absolute or relative) path of the json file containing the configuration. If `config_path` is not given or `No

Print the model structure. Args: container (keras.Model): The model or layer to be printed. max_lines (int): The maximum number o

Quantize and dequantize a tensor. Args: x (Tensor): The input tensor to be quantized and dequantized. min_x (Tensor, optional

Do quantization.

Run both the reference and the quantized versions of this module, and compare the outputs on every test vector. First the measure vectors are

Save configuration to a temporary json file.

Set the attribute data type.

Set the node's attr which data type is float.

Set the evaluation function registry. Args: eval_fns (Optional[Union[Callable, Dict, List[Dict]]]): The evaluation function(s) to

Initialize the AutoAlpha tuner with necessary parameters and components.

Initialization.

Create a singleton _Logger instance.

Absorb the scale to the layer at the output channel. Args: layer_name (str): The module name. scale (Tensor): The

Builds constant nodes needed for quantization of inputs.

Get the allowed dtype value of the node.