Custom Precision Quantization by Layer Name

custom_precision_quantization_by_layer_name(self, input_model_path: str, output_dir: str, dataset_path: str | None, precision_by_layer_name: List[PrecisionByLayer], default_weight_precision: QuantizationPrecision = QuantizationPrecision.INT8, default_activation_precision: QuantizationPrecision = QuantizationPrecision.INT8, metric: SimilarityMetric = SimilarityMetric.SNR, input_layers: List[Dict[str, int]] | None = None, wait_until_done: bool = True, sleep_interval: int = 30) → QuantizerMetadata

Apply custom precision quantization to a model, specifying precision for each layer name.

This function allows precise control over the quantization process by enabling the user to specify quantization precision (e.g., INT8, FP16) for each named layer within the model. The precision_by_layer_name parameter provides a list where each item details the target precision for a specific layer name, enabling customized quantization that can enhance model performance or compatibility.

Users can target specific layers to be quantized to lower precision for optimized model size and performance while keeping critical layers at higher precision for accuracy. Layers not explicitly listed in precision_by_layer_name will use default_weight_precision and default_activation_precision.

Parameters:

  • input_model_path (str) – The file path where the model is located.

  • output_dir (str) – The local folder path to save the quantized model.

  • dataset_path (str) – Path to the dataset. Useful for certain quantizations.

  • precision_by_layer_name (List[PrecisionByLayer]) – List of PrecisionByLayer objects that specify the desired precision for each layer name in the model. Each entry includes: name (str): The layer name (e.g., /backbone/conv_first/block/act/Mul_output_0). precision (QuantizationPrecision): The quantization precision level.

  • default_weight_precision (QuantizationPrecision) – Weight precision.

  • default_activation_precision (QuantizationPrecision) – Activation precision.

  • metric (SimilarityMetric) – Quantization quality metrics.

  • input_layers (List[InputShape], optional) – Target input shape for quantization (e.g., dynamic batch to static batch).

  • wait_until_done (bool) – If True, wait for the quantization result before returning the function. If False, request the quantization and return immediately.

  • sleep_interval (int) – Interval in seconds between checks when wait_until_done is True.

Raises:

e – If an error occurs during the model quantization.

Returns:

Quantization metadata containing status, paths, etc.

Return type:

QuantizerMetadata

Example

from netspresso import NetsPresso
from netspresso.enums import QuantizationPrecision


netspresso = NetsPresso(email="YOUR_EMAIL", password="YOUR_PASSWORD")

quantizer = netspresso.quantizer()

recommendation_metadata = quantizer.get_recommendation_precision(
    input_model_path="./examples/sample_models/test.onnx",
    output_dir="./outputs/quantized/automatic_quantization",
    dataset_path="./examples/sample_datasets/pickle_calibration_dataset_128x128.npy",
    weight_precision=QuantizationPrecision.INT8,
    activation_precision=QuantizationPrecision.INT8,
    threshold=0,
)
recommendation_precisions = quantizer.load_recommendation_precision_result(recommendation_metadata.recommendation_result_path)

quantization_result = quantizer.custom_precision_quantization_by_layer_name(
    input_model_path="./examples/sample_models/test.onnx",
    output_dir="./outputs/quantized/custom_precision_quantization_by_layer_name",
    dataset_path="./examples/sample_datasets/pickle_calibration_dataset_128x128.npy",
    precision_by_layer_name=recommendation_precisions.layers,
)