qbcompiler.configs.models.CalibrationConfig Class Reference

qbcompiler.configs.models.CalibrationConfig Class Reference#

Mobilint SDK qb Compiler: qbcompiler.configs.models.CalibrationConfig Class Reference
Mobilint SDK qb Compiler v1.2
MCS002-KR
qbcompiler.configs.models.CalibrationConfig Class Reference

Configuration for calibration during quantization. More...

Inheritance diagram for qbcompiler.configs.models.CalibrationConfig:

Classes

class  FastDist
class  Histogram
class  LayerOverrides
class  MaxPercentile
class  Statistics

Public Member Functions

"CalibrationConfig" with_updates (self, **kwargs)
 Return a copy with updated fields.

Static Public Attributes

 model_config
List method_list
int method = Field(default=1, alias="method")
List output_list
int output = Field(default=0, alias="output")
List mode_list
int mode = Field(default=1, alias="mode")
List clustering_methods_list
List clustering_methods = Field(default=[0], alias="clusteringMethods")
float act_scale_min
float act16_scale_min
float weight_scale_min
float weight16_scale_min
float min_clip_ratio
int max_calib_data_size
 MaxPercentile
 default_factory
 alias
 FastDist
 Histogram
 LayerOverrides
 Statistics

Detailed Description

Configuration for calibration during quantization.

Defines calibration and quantization parameterization used to derive activation/weight scales and related statistics during quantized compilation.

Parameters
methodint. Calibration method index:
0: WChALayer - Weight per-channel, Activation per-layer.
1: WChAMulti - Weight per-channel, Activation multi-layer.
2: WChALayerZeropoint - Weight per-channel, Activation per-layer with zeropoint.
3: WChAMultiZeropoint - Weight per-channel, Activation multi-layer with zeropoint.
outputint. Output quantization type index:
0: Layer - Per-layer quantization.
1: Ch - Per-channel quantization.
2: Sigmoid - Sigmoid-based quantization.
modeint. Quantization mode index:
0: Max - Maximum value calibration.
1: MaxPercentile - Maximum percentile calibration.
2: Histogram - Histogram-based calibration.
clustering_methodsint. LUT clustering method:
0: Scale - Scale-based 1D clustering.
1: Scale2D - Scale-based 2D clustering.
2: ReparameterizedL2 - Reparameterized L2 distance clustering.
3: SubgraphIoU - Subgraph intersection-over-union clustering.
4: DomainAwareL2 - Domain-aware L2 distance on common absolute grid.
5: OverlapWeightedL2 - L2 weighted by domain overlap (IoU).
6: Fast - Minimal search space for fast LUT optimization.
act_scale_minfloat. Minimum allowed activation scale (lower bound clamp)
act16_scale_minfloat. Minimum 16-bit activation scale (actScaleMin / 256)
weight_scale_minfloat. Minimum allowed weight scale (lower bound clamp)
weight16_scale_minfloat. Minimum 16-bit weight scale (weightScaleMin / 256)
min_clip_ratiofloat. Minimum clip ratio constraint applied during calibration
max_calib_data_sizeint. Maximum number of calibration samples kept after loading or generation
max_percentileMaxPercentile. MaxPercentile mode configuration
fast_distFastDist. Fast distribution calibration configuration
histogramHistogram. Histogram-based calibration configuration
layer_overridesLayerOverrides. Layer-specific override settings for calibration
statisticsStatistics. Statistics save/load configuration with percentile selection

Definition at line 141 of file models.py.

Member Function Documentation

◆ with_updates()

"CalibrationConfig" qbcompiler.configs.models.CalibrationConfig.with_updates ( self,
** kwargs )

Return a copy with updated fields.

Definition at line 437 of file models.py.

Member Data Documentation

◆ model_config

qbcompiler.configs.models.CalibrationConfig.model_config
static
Initial value:
= ConfigDict(
populate_by_name=True,
extra="forbid",
)

Definition at line 182 of file models.py.

◆ method_list

List qbcompiler.configs.models.CalibrationConfig.method_list
static
Initial value:
= Field(
default=["WChALayer", "WChAMulti", "WChALayerZeropoint", "WChAMultiZeropoint"],
alias="methodList",
)

Definition at line 187 of file models.py.

◆ method

int qbcompiler.configs.models.CalibrationConfig.method = Field(default=1, alias="method")
static

Definition at line 191 of file models.py.

◆ output_list

List qbcompiler.configs.models.CalibrationConfig.output_list
static
Initial value:
= Field(
default=["Layer", "Ch", "Sigmoid"], alias="outputList"
)

Definition at line 192 of file models.py.

◆ output

int qbcompiler.configs.models.CalibrationConfig.output = Field(default=0, alias="output")
static

Definition at line 195 of file models.py.

◆ mode_list

List qbcompiler.configs.models.CalibrationConfig.mode_list
static
Initial value:
= Field(
default=["Max", "MaxPercentile", "Histogram"], alias="modeList"
)

Definition at line 196 of file models.py.

◆ mode

int qbcompiler.configs.models.CalibrationConfig.mode = Field(default=1, alias="mode")
static

Definition at line 199 of file models.py.

◆ clustering_methods_list

List qbcompiler.configs.models.CalibrationConfig.clustering_methods_list
static
Initial value:
= Field(
default=[
"Scale",
"Scale2D",
"ReparameterizedL2",
"SubgraphIoU",
"DomainAwareL2",
"OverlapWeightedL2",
"Fast",
],
alias="clusteringMethodsList",
)

Definition at line 200 of file models.py.

◆ clustering_methods

List qbcompiler.configs.models.CalibrationConfig.clustering_methods = Field(default=[0], alias="clusteringMethods")
static

Definition at line 212 of file models.py.

◆ act_scale_min

float qbcompiler.configs.models.CalibrationConfig.act_scale_min
static
Initial value:
= Field(
default=0.0005,
alias="actScaleMin",
description="Minimum allowed activation scale (lower bound clamp)",
ge=0,
le=1,
)

Definition at line 214 of file models.py.

◆ act16_scale_min

float qbcompiler.configs.models.CalibrationConfig.act16_scale_min
static
Initial value:
= Field(
default=1.953125e-06,
alias="act16ScaleMin",
description="Minimum 16-bit activation scale (actScaleMin / 256)",
)

Definition at line 221 of file models.py.

◆ weight_scale_min

float qbcompiler.configs.models.CalibrationConfig.weight_scale_min
static
Initial value:
= Field(
default=1e-06,
alias="weightScaleMin",
description="Minimum allowed weight scale (lower bound clamp)",
ge=0,
le=1,
)

Definition at line 226 of file models.py.

◆ weight16_scale_min

float qbcompiler.configs.models.CalibrationConfig.weight16_scale_min
static
Initial value:
= Field(
default=3.90625e-09,
alias="weight16ScaleMin",
description="Minimum 16-bit weight scale (weightScaleMin / 256)",
)

Definition at line 233 of file models.py.

◆ min_clip_ratio

float qbcompiler.configs.models.CalibrationConfig.min_clip_ratio
static
Initial value:
= Field(
default=-1,
alias="minClipRatio",
description="Minimum clip ratio constraint applied during calibration",
ge=-1,
le=1,
)

Definition at line 238 of file models.py.

◆ max_calib_data_size

int qbcompiler.configs.models.CalibrationConfig.max_calib_data_size
static
Initial value:
= Field(
default=-1,
alias="maxCalibDataSize",
description="Maximum number of calibration samples kept after loading or generation",
ge=-1,
)

Definition at line 245 of file models.py.

◆ MaxPercentile

qbcompiler.configs.models.CalibrationConfig.MaxPercentile
static

Definition at line 289 of file models.py.

◆ default_factory

qbcompiler.configs.models.CalibrationConfig.default_factory
static

Definition at line 290 of file models.py.

◆ alias

qbcompiler.configs.models.CalibrationConfig.alias
static

Definition at line 290 of file models.py.

◆ FastDist

qbcompiler.configs.models.CalibrationConfig.FastDist
static

Definition at line 307 of file models.py.

◆ Histogram

qbcompiler.configs.models.CalibrationConfig.Histogram
static

Definition at line 373 of file models.py.

◆ LayerOverrides

qbcompiler.configs.models.CalibrationConfig.LayerOverrides
static

Definition at line 398 of file models.py.

◆ Statistics

qbcompiler.configs.models.CalibrationConfig.Statistics
static

Definition at line 435 of file models.py.


The documentation for this class was generated from the following file: