Post-Training Compression

coremltools.optimize.coreml.linear_quantize_weights(*args, **kwargs)

Utility function to convert a float precision MLModel of type mlprogram, which uses float-precision weights, into a compressed MLModel that uses n-bit weights (currently only support n=4 and n=8). This is achieved by converting the float weight values that are stored in the const op into the constexpr_affine_dequantize or constexpr_blockwise_shift_scale op (based on model’s minimum deployment target).

This function uses linear quantization on the float weights, providing up to 4x (for 4-bit) savings in storage compared to float 16, or up to 4x savings compared to float 32. All computation at runtime uses float precision; the precision of the intermediate tensors and the compute precision of the ops are not altered.

For each weight, this utility function converts the weight into the int4/8 or uint4/8 type using either linear interpolation ("linear" mode) or linear symmetric interpolation ("linear_symmetric" mode, the default).

Linear interpolation

The following description uses 8-bit quantization to illustrate, and 4-bit is similar to it.

Linear interpolation ("linear" mode) maps the min/max of the float range to the 8-bit integer range [low, high] using a zero point (also called quantization bias, or offset) and a scale factor. For the int8 quantization, [low, high] = [-128, 127], while uint8 quantization uses range [0, 255].

"linear" mode uses the quantization formula:

\[w_r = s * (w_q - z)\]

Where:

• \(w_r\) and \(s\) are of type float.

• \(w_r`\) represents the float precision weight.

• \(s\) represents the scale.

• \(w_q\) and \(z\) are of type 8-bit integer.

• \(w_q\) represents quantized weight.

• \(z\) represents the zero point.

Quantized weights are computed as follows:

\[w_q = cast\_to\_8\_bit\_integer(w_r / s + cast\_to\_float(z))\]

Note: \(cast\_to\_8\_bit\_integer\) is the process of clipping the input to range [low, high] followed by rounding and casting to 8-bit integer.

In "linear" mode, s, z are computed by mapping the original float range [A, B] into the 8-bit integer range [-128, 127] or [0, 255]. That is, you are solving the following linear equations:

• B = s * (high - z)

• A = s * (low - z)

The equations result in the following:

• s = (B - A) / (high - low)

• z = cast_to_8_bit_integer((low * B - high * A) / (B - A))

When the rank of weight w is 1, then s and z are both scalars. When the rank of the weight is greater than 1, then s and z are both vectors. In that case, scales are computed per channel, in which channel is the output dimension, which corresponds to the first dimension for ops such as conv and linear, and the second dimension for the conv_transpose op.

For "linear" mode, \(A = min(w_r)\), \(B = max(w_r)\).

Linear symmetric interpolation

With linear symmetric interpolation ("linear_symmetric" mode, the default), rather than mapping the exact min/max of the float range to the quantized range, the function chooses the maximum absolute value between the min/max, which results in a floating-point range that is symmetric with respect to zero. This also makes the resulting zero point 0 for int8 weight and 127 for uint8 weight.

For "linear_symmetric" mode:

• \(A = -R\) and \(B = R\), where \(R = max(abs(w_r))\).

• This function maps to the range of [-127, 127] for int8 weight and [0, 254] for uint8 weight.

• The result is s=(B-A)/254 -> s=2R/254 -> s=R/127.

• Solving for z:

  • int8: z = (-127 * R + 127 * R)/2R -> z=0.

  • uint8: z = (0 * R + 254 * R)/2R -> z=127.

Parameters:

mlmodel: MLModel

Model to be quantized. This MLModel should be of type mlprogram.

config: OptimizationConfig

An OptimizationConfig object that specifies the parameters for weight quantization.

joint_compression: bool

Specification of whether or not to further compress the already-compressed input MLModel to a jointly compressed MLModel. See the blockwise_palettize_weights graph pass for information about which compression schemas could be further jointly palettized.

Take “palettize + quantize” as an example of joint compression, where the input MLModel is already palettized, and the palettization’s lookup table will be further quantized. In such an example, the weight values are represented by constexpr_blockwise_shift_scale + constexpr_lut_to_dense ops: lut(int8) -> constexpr_blockwise_shift_scale -> lut(fp16) -> constexpr_lut_to_dense -> dense(fp16)

Returns:

model: MLModel

The quantized MLModel instance.

Examples

import coremltools as ct
import coremltools.optimize as cto

model = ct.coreml.models.MLModel("my_model.mlpackage")
config = cto.coreml.OptimizationConfig(
    global_config=cto.coreml.OpLinearQuantizerConfig(mode="linear_symmetric")
)
compressed_model = cto.coreml.linear_quantize_weights(model, config)

coremltools.optimize.coreml.prune_weights(*args, **kwargs)

Utility function to convert a float precision MLModel of type mlprogram to a compressed MLModel using sparse representation. The const ops storing weight values are replaced by constexpr_sparse_to_dense ops.

This function is useful if the model is trained with pruning techniques so that a lot of weights have zero values. If a large percentage of weight values are zero, a sparse representation is more efficient than a dense one (the default).

The sparsified weights are stored in a bit mask. If the weight values are {0, 0, 0, 0, 0, 0, 0, 56.3}, its sparse representation contains a bit mask with ones on locations where the value is non-zero: 00000001b. This is accompanied by non-zero data, which is a size-1 vector of value {56.3}.

For example, given the following:

• weight = [0.3, 0, 0, 0.5, 0, 0]

• non_zero_data, bit_mask = sparsify(weight)

The indices of the non-zero elements are:

• non_zero_data = [0.3, 0.5]

• bit_mask = "100100"

Parameters:

mlmodel: MLModel

Model to be sparsified. This MLModel should be of type mlprogram.

config: OptimizationConfig

An OptimizationConfig object that specifies the parameters for weight pruning.

joint_compression: bool

Specification of whether or not to further prune the already-compressed input MLModel to a jointly compressed MLModel. See the prune_weights graph pass for information about which compression schemas could be further pruned.

Take “quantize + prune” as an example of joint compression, where the input MLModel is already quantized, and it will be further pruned. In such an example, the weight values are represented by constexpr_sparse_blockwise_shift_scale + constexpr_sparse_to_dense ops: quantized(sparse) -> constexpr_sparse_blockwise_shift_scale -> weight(sparse) -> constexpr_sparse_to_dense -> weight(dense)

Returns:

model: MLModel

The sparse MLModel instance.

Examples

import coremltools as ct
import coremltools.optimize as cto

model = ct.models.MLModel("my_model.mlpackage")
config = cto.coreml.OptimizationConfig(
    global_config=cto.coreml.OpThresholdPrunerConfig(threshold=1e-12)
)
compressed_model = cto.coreml.prune_weights(model, config)

coremltools.optimize.coreml.palettize_weights(*args, **kwargs)

Utility function to convert a float precision MLModel of type mlprogram to a compressed MLModel by reducing the overall number of weights using one or more lookup tables (LUT). A LUT contains a list of float values. An n-bit LUT has \(2^{n-bits}\) entries.

For example, a float weight vector such as {0.3, 0.3, 0.5, 0.5} can be compressed using a 1-bit LUT: {0.3, 0.5}. In this case the float vector can be replaced with a 1-bit vector {0, 0, 1, 1}.

This function iterates over all the weights in the mlprogram, discretizes its values, and constructs the LUT according to the algorithm specified in mode. The float values are then converted to the n-bit values, and the LUT is saved alongside each weight. The const ops storing weight values are replaced by constexpr_lut_to_dense ops.

At runtime, the LUT and the n-bit values are used to reconstruct the float weight values, which are then used to perform the float operation the weight is feeding into.

Consider the following example of "uniform" mode (a linear histogram):

• nbits = 4

• mode = "uniform"

• weight = [0.11, 0.19, 0.3, 0.08, 0.0, 0.02]

The weight can be converted to a palette with indices [0, 1, 2, 3] (2 bits). The indices are a byte array.

The data range [0.0, 0.3] is divided into four partitions linearly, which is [0.0, 0.1, 0.2, 0.3].

• The LUT would be [0.0, 0.1, 0.2, 0.3].

• The weight is rounded to [0.1, 0.2, 0.3, 0.1, 0.0, 0.0] and represented in the palette as indices [01b, 10b, 11b, 01b, 00b, 00b].

Parameters:

mlmodel: MLModel

Model to be converted by a LUT. This MLModel should be of type mlprogram.

config: OptimizationConfig

An OptimizationConfig object that specifies the parameters for weight palettization.

joint_compression: bool

Specification of whether or not to further compress the already-compressed input MLModel to a jointly compressed MLModel. See the channelwise_palettize_weights graph pass for information about which compression schemas could be further jointly palettized.

Take “prune + palettize” as an example of joint compression, where the input MLModel is already pruned, and the non-zero entries will be further palettized. In such an example, the weight values are represented by constexpr_lut_to_sparse + constexpr_sparse_to_dense ops: lut(sparse) -> constexpr_lut_to_sparse -> weight(sparse) -> constexpr_sparse_to_dense -> weight(dense)

Returns:

model: MLModel

The palettized MLModel instance.

Examples

import coremltools as ct
import coremltools.optimize as cto

model = ct.models.MLModel("my_model.mlpackage")
config = cto.coreml.OptimizationConfig(
    global_config=cto.coreml.OpPalettizerConfig(mode="kmeans", nbits=4)
)
compressed_model = cto.coreml.palettize_weights(model, config)

coremltools.optimize.coreml.decompress_weights(*args, **kwargs)

Utility function to convert weights that are sparse or palettized or affine quantized, back to the float format. That is, convert any of the following three ops to mb.const:

1. constexpr_affine_dequantize

2. constexpr_lut_to_dense

3. constexpr_sparse_to_dense

Parameters:

mlmodel: MLModel

Model which will be decompressed.

Returns:

model: MLModel

The MLModel with no constexpr ops included.

Examples

import coremltools as ct

model = ct.models.MLModel("my_compressed_model.mlpackage")
decompressed_model = ct.optimize.coreml.decompress_weights(model)

coremltools.optimize.coreml.get_weights_metadata(*args, **kwargs)

Utility function to get the weights metadata as a dictionary, which maps the weight’s name to its corresponding CoreMLWeightMetaData.

CoreMLWeightMetaData contains the following attributes:

1. val: The weight data.

2. sparsity: the percentile of the element whose absolute value <= 1e-12.

3. unique_values: number of unique values in the weight.

4. child_ops: meta information of the child ops in which the weight is feeding into.

Parameters:

mlmodel: MLModel

Model in which the weight metadata is retrieved from.

weight_threshold: int

• The size threshold, above which weights are returned. That is, a weight tensor is included in the resulting dictionary only if its total number of elements are greater than weight_threshold. For example, if weight_threshold = 1024 and a weight tensor is of shape [10, 20, 1, 1], hence 200 elements, it will not be returned by the get_weights_metadata API.

• If not provided, it will be set to 2048, in which weights bigger than 2048 elements are returned.

Returns:

dict[str, CoreMLWeightMetaData]

A dict that maps weight’s name to its metadata.

Examples

In this example, there are two weights whose sizes are greater than 2048. A weight named conv_1_weight is feeding into a conv op named conv_1, while another weight named linear_1_weight is feeding into a linear op named linear_1. You can access the metadata by weight_metadata_dict["conv_1_weight"], and so on.

import coremltools as ct

mlmodel = ct.models.MLModel("my_model.mlpackage")
weight_metadata_dict = ct.optimize.coreml.get_weights_metadata(
    mlmodel, weight_threshold=2048
)

# get the weight names with size > 25600
large_weights = []
for k, v in weight_metadata_dict.items():
    if v.val.size >= 25600:
        large_weights.append(k)

# get the weight names with sparsity >= 50%
sparse_weights = []
for k, v in weight_metadata_dict.items():
    if v.sparsity >= 0.5:
        sparse_weights.append(k)

# get the weight names with unique elements <= 16
palettized_weights = []
for k, v in weight_metadata_dict.items():
    if v.unique_values <= 16:
        palettized_weights.append(k)

# print out the dictionary
print(weight_metadata_dict)

The output from the above example would be:

conv_1_weight
[
    val: np.ndarray(shape=(32, 64, 2, 2), dtype=float32)
    sparsity: 0.5
    unique_values: 4097
    child_ops: [
        conv(name=conv_1, weight=conv_1_weight, ...)
    ]
]
linear_1_weight
[
    val: np.ndarray(shape=(128, 64), dtype=float32)
    sparsity: 0.2501220703125
    unique_values: 4
    child_ops: [
        linear(name=linear_1, weight=linear_1_weight, ...)
    ]
]

class coremltools.optimize.coreml.CoreMLWeightMetaData(val: ndarray, sparsity: float | None = NOTHING, unique_values: int | None = NOTHING, child_ops: List[CoreMLOpMetaData] | None = None)

A container class that stores weight meta data.

The class has the following attributes:

Parameters:

val: numpy.ndarray

The weight data.

sparsity: float

The percentile of the element whose absolute value <= 1e-12.

unique_values: int

Number of unique values in the weight.

child_ops: list[CoreMLOpMetaData]

A list of CoreMLOpMetaData which contains information of child ops in which the weight is feeding into.

The attributes can be accessed by: child_ops[idx].op_type: The operation type of the idx ‘th child op. child_ops[idx].name: The name of the idx ‘th child op.

Other op-dependant attributes also can be accessed. For instance, if idx ‘th child op is a conv layer, child_ops[idx].weight will return its weight name.

For more details, please refer to the CoreMLOpMetaData doc string.

Examples

import numpy as np
from coremltools.optimize.coreml import CoreMLWeightMetaData

data = np.array([[1.0, 0.0], [0.0, 6.0]], dtype=np.float32)
meta_data = CoreMLWeightMetaData(data)
print(meta_data)

Outputs:

[
    val: np.ndarray(shape=(2, 2), dtype=float32)
    sparsity: 0.5
    unique_values: 3
]

class coremltools.optimize.coreml.CoreMLOpMetaData(op_type: str, name: str, params_name_mapping: Dict[str, str])

A container class that stores op meta data.

The class has the following attributes:

Parameters:

op_type: str

The type of the op. For instance: conv, linear, and so on.

name: str

The name of the op.

params_name_mapping: dict[str, str]

A dict that maps the op’s constant parameters to its corresponding weight name. For instance, given a conv op with params_name_mapping,

{
    "weight": "conv_1_weight",
    "bias": "conv_1_bias",
}

means that the weight and bias of this op are named conv_1_weight, conv_1_bias, respectively.