Update the state that need to be evaluated. Args: preds: The prediction result. labels: The reference. Defaults to No

Update the state that need to be evaluated. Args: preds: The prediction result. labels: The reference. Defaults to No

Update the state that need to be evaluated. Args: preds: The prediction result. labels: The reference. Defaults to No

Update the state that need to be evaluated. Args: preds: The prediction result. labels: The reference. Defaults to No

Update ipex_config.json with smoothquant scale info generated by our algorithm. Args: ipex_config_path (str): a path to temporary ipex

Validate and inference the input and output tensor names of graph_def. Args: graph_def (tf.compat.v1.GraphDef): tf.compat.v1.GraphDef obj

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

Return weight_name_mapping function.

Compute detection metrics. Args: include_metrics_per_category: Whether include metrics per category. all_metrics_per_cate

Compute detection metrics. Args: include_metrics_per_category: Whether include metrics per category. all_metrics_per_cate

Compute detection metrics. Args: include_metrics_per_category: Whether include metrics per category. all_metrics_per_cate

Export detections of a single image to COCO format. This function converts detections represented as numpy arrays to dictionaries that can be

Export detections of a single image to COCO format. This function converts detections represented as numpy arrays to dictionaries that can be

Export detections of a single image to COCO format. This function converts detections represented as numpy arrays to dictionaries that can be

Export groundtruth of a single image to COCO format. This function converts groundtruth detection annotations represented as numpy arrays to

Export groundtruth of a single image to COCO format. This function converts groundtruth detection annotations represented as numpy arrays to

Export groundtruth of a single image to COCO format. This function converts groundtruth detection annotations represented as numpy arrays to

Return whether COCO Eval is configured to evaluate in agnostic mode.

Return whether COCO Eval is configured to evaluate in agnostic mode.

Return whether COCO Eval is configured to evaluate in agnostic mode.

Fetch dictionary holding category information given category id. Args: category_id: integer id Returns: dictiona

Fetch dictionary holding category information given category id. Args: category_id: integer id Returns: dictiona

Fetch dictionary holding category information given category id. Args: category_id: integer id Returns: dictiona

Return the list of IDs of all valid categories.

Return the list of IDs of all valid categories.

Return the list of IDs of all valid categories.

Load annotations dictionary into COCO datastructure. See http://mscoco.org/dataset/#format for a description of the annotations forma

Load annotations dictionary into COCO datastructure. See http://mscoco.org/dataset/#format for a description of the annotations forma

Load annotations dictionary into COCO datastructure. See http://mscoco.org/dataset/#format for a description of the annotations forma

The __call__ function of custom quantized AveragePooling2D layer.

Initialization function.

Init the GraphTrace object.

Init Static quantization config. Args: weight_dtype (str): Data type for weights, default is "fp8_e4m3". activation_d

Initialize the min/max observer layer. Args: *args: Positional arguments for the base layer. **kwargs: Keyword argume

Init static quantization config. Args: weight_dtype (str): Data type for weights, default is "int". weight_sym (bool)

Initialization. Args: local_config_file: local configuration file name. performance_only: oob performance only mode.

Initialize custom quantized AveragePooling2D layer.

Construct a COCOWrapper. See http://mscoco.org/dataset/#format for a description of the format. By default, the coco.COCO class const

Construct a COCOWrapper. See http://mscoco.org/dataset/#format for a description of the format. By default, the coco.COCO class const

Construct a COCOWrapper. See http://mscoco.org/dataset/#format for a description of the format. By default, the coco.COCO class const

Initialize the score list.

Return the number of batches.

Bypass BaseModel __new__ to allow manual initialization. Args: *args: Positional arguments for object creation. **kwa

Override the setattr function to propagate updates.

Absorb the scale to the layer at output channel. Args: layer: the module. scale: the scale to be absorbed.

Insert observers into the block modules. Args: block_modules (list): The block modules to which the observer will be inserted.

Insert observers into the given modules. Args: modules (list): The modules to which the observer will be inserted.

Set quantized pooling node attributes.

Adjust the weights and biases. Args: absorb_to_layer (dict): A dictionary mapping absorb layers to smooth quantized layers.

In-place adjust weight by scale. Args: scale: smooth scale with the shape (ic,) weight_node: reference to the origina

Analyze all the ConcatV2 nodes recursively.

Quantize ConcatV2 if it's quantizable.

Quantize ConcatV2.

Quantize ConcatV2.

Quantize AvgPool/MaxPool.

Quantize AvgPool/MaxPool. Dequantize + MaxPool + QuantizeV2 Dequantize + MaxPool3D + QuantizeV2 Dequantize + AvgPool + Quanti

In quantized mode, assert the reference model is not quantized, and the quantized model is. Otherwise, assert that both are not quantized.

Perform alpha-tuning to obtain layer-wise optimal alpha values and adjust parameters accordingly.

Perform blockwise-alpha-tuning to obtain optimal alpha values and adjust parameters accordingly.

BF16 conversion for the model. Args: bf16_node_name: nodes converted to BF16 op list

Reconstructs layers' weights by building each layer from its saved input shape and then assigning the original variable values.

Replace quantized Linear with HQQLinear.

Replaces original add_weight() (used internally by from_config()) with faster version, which creates copies of original weights. Then calls f

Build lookup for synset to human-readable label. Args: imagenet_metadata_file: string, path to file containing mapping from synset to hum

Builds a path->value map for name-based copy in add_weight. Keys use the path with the top-level model name stripped, because from_config reb

Scale propagation pattern match and create new const nodes.

Calculate the adjustment scales. Args: absorb_to_layer (dict): A dictionary mapping absorb layers to smooth quantized layers.

Calculate scale and zero-point for quantization. Args: input_scale (Tensor): The scale for input features. input_

Apply calibration. Args: model(tf.keras.Model): The model inserted with FakeQuant layers for calibration. dataloader(

Check model's arguments.

Checks if the given config matches all specified criteria for weight and activation (scale_value_type, rounding_method, granularity). Returns

Check if this is the 'max context' doc span for the token.

Verify that Intel Extension for TensorFlow is installed.

Decide which method will be used to get graph from the saved_model . Args: model (string or tf.keras.Model): model path or tf.keras.Model

Check if calibration is needed. Args: alpha (float or dict): Current alpha values. percentile (float): Current pe

Check if the node_type in the allowed op list.

The function that checks format for conv ops.

Check if the installed tensorflow version is supported.

Remove group convolution layers except depthwise convolution. Args: module (torch.nn.Module): The module to process.

Remove group conv except depthwise conv.

Performs invalid character removal and whitespace cleanup on text.

Compute softmax probability over raw logits.