Mod Configuration

Mod Configuration#

Mobilint SDK qb Compiler: Mod Configuration

Mobilint SDK qb Compiler v0.12.0.0

MCS002-EN

Minimum Output Difference configuration definitions used by Mobilint compiler for model quantization. More...

Classes
class	qubee.configs.mod_config.ModConfig
	Configuration for Minimum Output Difference algorithm. More...

Functions
	qubee.configs.mod_config.ModConfig.__init__ (self, bool apply, int epochs, int warmup_epochs, numbers.Number lr_act_scale, numbers.Number lr_zeropoint, numbers.Number lr_weight_scale, numbers.Number lr_weight, numbers.Number lr_bias, numbers.Number lr_min_ratio, int batch_size, numbers.Number q_drop, bool quantize_weight, str weight_scale_init, str downresol_mode, str post_processing, numbers.Number box_conf_thres, numbers.Number box_iou_thres, str loss_type, bool use_outputs, numbers.Number kl_temperature, numbers.Number recon_prob, numbers.Number recon_coeff, numbers.Number lambda_0, numbers.Number lambda_1, numbers.Number lambda_2, List anchors, bool use_xyxy, str save_dir, int seed, List[str] apply_layers, List[str] exclude_layers, bool optional=optional)
	Initialize the ModConfig.
	qubee.configs.mod_config.ModConfig.default_config (cls)
	qubee.configs.mod_config.ModConfig.from_dict (cls, dict config_dict)
	Create ModConfig from dictionary (JSON structure)
	qubee.configs.mod_config.ModConfig.to_dict (self)
ModConfig	qubee.configs.mod_config.get_mod_config (**kwargs)
	Create ModConfig with partial parameters merged with defaults.

Variables
bool	qubee.configs.mod_config.ModConfig.optional = True
dict	qubee.configs.mod_config.ModConfig.DEFAULTS
dict	qubee.configs.mod_config.ModConfig.py_to_cpp_name
dict	qubee.configs.mod_config.ModConfig.min_max_check
	qubee.configs.mod_config.ModConfig.apply = apply
	qubee.configs.mod_config.ModConfig.epochs = epochs
	qubee.configs.mod_config.ModConfig.warmup_epochs = warmup_epochs
	qubee.configs.mod_config.ModConfig.lr_act_scale = lr_act_scale
	qubee.configs.mod_config.ModConfig.lr_zeropoint = lr_zeropoint
	qubee.configs.mod_config.ModConfig.lr_weight_scale = lr_weight_scale
	qubee.configs.mod_config.ModConfig.lr_weight = lr_weight
	qubee.configs.mod_config.ModConfig.lr_bias = lr_bias
	qubee.configs.mod_config.ModConfig.lr_min_ratio = lr_min_ratio
	qubee.configs.mod_config.ModConfig.batch_size = batch_size
	qubee.configs.mod_config.ModConfig.q_drop = q_drop
	qubee.configs.mod_config.ModConfig.quantize_weight = quantize_weight
	qubee.configs.mod_config.ModConfig.weight_scale_init = weight_scale_init
	qubee.configs.mod_config.ModConfig.downresol_mode = downresol_mode
	qubee.configs.mod_config.ModConfig.post_processing = post_processing
	qubee.configs.mod_config.ModConfig.box_conf_thres = box_conf_thres
	qubee.configs.mod_config.ModConfig.box_iou_thres = box_iou_thres
	qubee.configs.mod_config.ModConfig.loss_type = loss_type
	qubee.configs.mod_config.ModConfig.use_outputs = use_outputs
	qubee.configs.mod_config.ModConfig.kl_temperature = kl_temperature
	qubee.configs.mod_config.ModConfig.recon_prob = recon_prob
	qubee.configs.mod_config.ModConfig.recon_coeff = recon_coeff
	qubee.configs.mod_config.ModConfig.lambda_0 = lambda_0
	qubee.configs.mod_config.ModConfig.lambda_1 = lambda_1
	qubee.configs.mod_config.ModConfig.lambda_2 = lambda_2
	qubee.configs.mod_config.ModConfig.anchors = anchors
	qubee.configs.mod_config.ModConfig.use_xyxy = use_xyxy
	qubee.configs.mod_config.ModConfig.save_dir = save_dir
	qubee.configs.mod_config.ModConfig.seed = seed
	qubee.configs.mod_config.ModConfig.apply_layers = apply_layers
	qubee.configs.mod_config.ModConfig.exclude_layers = exclude_layers

Detailed Description

Minimum Output Difference configuration definitions used by Mobilint compiler for model quantization.

Function Documentation

◆ init()

qubee.configs.mod_config.ModConfig.__init__	(		self,
		bool	apply,
		int	epochs,
		int	warmup_epochs,
		numbers.Number	lr_act_scale,
		numbers.Number	lr_zeropoint,
		numbers.Number	lr_weight_scale,
		numbers.Number	lr_weight,
		numbers.Number	lr_bias,
		numbers.Number	lr_min_ratio,
		int	batch_size,
		numbers.Number	q_drop,
		bool	quantize_weight,
		str	weight_scale_init,
		str	downresol_mode,
		str	post_processing,
		numbers.Number	box_conf_thres,
		numbers.Number	box_iou_thres,
		str	loss_type,
		bool	use_outputs,
		numbers.Number	kl_temperature,
		numbers.Number	recon_prob,
		numbers.Number	recon_coeff,
		numbers.Number	lambda_0,
		numbers.Number	lambda_1,
		numbers.Number	lambda_2,
		List	anchors,
		bool	use_xyxy,
		str	save_dir,
		int	seed,
		List[str]	apply_layers,
		List[str]	exclude_layers,
		bool	optional = optional )

Initialize the ModConfig.

Parameters

apply	bool. Whether to apply MOD.
epochs	int. Number of training epochs.
warmup_epochs	int. Number of warmup epochs.
lr_act_scale	float. Learning rate for activation scale.
lr_zeropoint	float. Learning rate for zeropoint.
lr_weight_scale	float. Learning rate for weight scale.
lr_weight	float. Learning rate for weight.
lr_bias	float. Learning rate for bias.
lr_min_ratio	float. Minimum learning rate ratio for scheduler.
batch_size	int. Batch size for training.
q_drop	float. QDrop ratio.
quantize_weight	bool. Quantize weight during training.
weight_scale_init	str. Weight scale initialization method.
downresol_mode	str. Down-resolution mode for fake quantization.
post_processing	str. Post-processing function.
box_conf_thres	float. Box confidence threshold for NMS.
box_iou_thres	float. Box IoU threshold for NMS.
loss_type	str. Loss function type.
use_outputs	bool. Compare only intermediate layers.
kl_temperature	float. Temperature for KL divergence loss.
recon_prob	float. Sampling probability for intermediate layers.
recon_coeff	float. Coefficient for intermediate layer loss.
lambda_0	float. First coefficient for AnchorlessDetection loss.
lambda_1	float. Second coefficient for AnchorlessDetection loss.
lambda_2	float. Third coefficient for AnchorlessDetection loss.
anchors	List. Anchor boxes configuration.
use_xyxy	bool. Use XYXY format.
save_dir	str. Directory to save results.
seed	int. Random seed.
apply_layers	List[str]. Layers to apply MOD.
exclude_layers	List[str]. Layers to exclude from MOD.
optional	bool. Indicates whether this config is optional. Use the default value defined as a class variable.

Definition at line 114 of file mod_config.py.

◆ default_config()

qubee.configs.mod_config.ModConfig.default_config ( cls )

Definition at line 221 of file mod_config.py.

◆ from_dict()

qubee.configs.mod_config.ModConfig.from_dict	(		cls,
		dict	config_dict )

Create ModConfig from dictionary (JSON structure)

Definition at line 225 of file mod_config.py.

◆ to_dict()

qubee.configs.mod_config.ModConfig.to_dict ( self )

Definition at line 310 of file mod_config.py.

◆ get_mod_config()

ModConfig qubee.configs.mod_config.get_mod_config ( ** kwargs )

Create ModConfig with partial parameters merged with defaults.

Definition at line 356 of file mod_config.py.

Variable Documentation

◆ optional

bool qubee.configs.mod_config.ModConfig.optional = True

static

Definition at line 21 of file mod_config.py.

◆ DEFAULTS

qubee.configs.mod_config.ModConfig.DEFAULTS

static

Definition at line 22 of file mod_config.py.

◆ py_to_cpp_name

dict qubee.configs.mod_config.ModConfig.py_to_cpp_name

static

Initial value:

=  {
        "apply": "apply",
        "epochs": "epochs",
        "warmup_epochs": "warmupEpochs",
        "lr_min_ratio": "lrMinRatio",
        "lr_act_scale": "actScaleLR",
        "lr_zeropoint": "zeropointLR",
        "lr_weight_scale": "weightScaleLR",
        "lr_weight": "weightLR",
        "lr_bias": "biasLR",
        "batch_size": "batchSize",
        "q_drop": "qDrop",
        "quantize_weight": "quantizeWeight",
        "weight_scale_init": "weightScaleInit",
        "downresol_mode": "downresolMode",
        "post_processing": "post",
        "box_conf_thres": "boxConfThres",
        "box_iou_thres": "boxIoUThres",
        "loss_type": "lossType",
        "use_outputs": "useOutputs",
        "kl_temperature": "KLTemperature",
        "recon_prob": "reconProb",
        "recon_coeff": "reconCoeff",
        "lambda_0": "lambda_0",
        "lambda_1": "lambda_1",
        "lambda_2": "lambda_2",
        "use_xyxy": "useXYXY",
        "save_dir": "saveDir",
        "seed": "seed",
    }

Definition at line 60 of file mod_config.py.

◆ min_max_check

dict qubee.configs.mod_config.ModConfig.min_max_check

static

Initial value:

=  {
        "epochs": [0, None],
        "warmup_epochs": [0, None],
        "lr_min_ratio": [0, None],
        "lr_act_scale": [0, None],
        "lr_zeropoint": [0, None],
        "lr_weight_scale": [0, None],
        "lr_weight": [0, None],
        "lr_bias": [0, None],
        "batch_size": [0, None],
        "q_drop": [0, 1],
        "box_conf_thres": [0, None],
        "box_iou_thres": [0, None],
        "kl_temperature": [0, None],
        "recon_prob": [0, 1],
        "recon_coeff": [0, None],
        "lambda_0": [0, None],
        "lambda_1": [0, None],
        "lambda_2": [0, None],
        "seed": [0, None],
    }

Definition at line 91 of file mod_config.py.

◆ apply

qubee.configs.mod_config.ModConfig.apply = apply

Definition at line 185 of file mod_config.py.

◆ epochs

qubee.configs.mod_config.ModConfig.epochs = epochs

Definition at line 186 of file mod_config.py.

◆ warmup_epochs

qubee.configs.mod_config.ModConfig.warmup_epochs = warmup_epochs

Definition at line 187 of file mod_config.py.

◆ lr_act_scale

qubee.configs.mod_config.ModConfig.lr_act_scale = lr_act_scale

Definition at line 188 of file mod_config.py.

◆ lr_zeropoint

qubee.configs.mod_config.ModConfig.lr_zeropoint = lr_zeropoint

Definition at line 189 of file mod_config.py.

◆ lr_weight_scale

qubee.configs.mod_config.ModConfig.lr_weight_scale = lr_weight_scale

Definition at line 190 of file mod_config.py.

◆ lr_weight

qubee.configs.mod_config.ModConfig.lr_weight = lr_weight

Definition at line 191 of file mod_config.py.

◆ lr_bias

qubee.configs.mod_config.ModConfig.lr_bias = lr_bias

Definition at line 192 of file mod_config.py.

◆ lr_min_ratio

qubee.configs.mod_config.ModConfig.lr_min_ratio = lr_min_ratio

Definition at line 193 of file mod_config.py.

◆ batch_size

qubee.configs.mod_config.ModConfig.batch_size = batch_size

Definition at line 194 of file mod_config.py.

◆ q_drop

qubee.configs.mod_config.ModConfig.q_drop = q_drop

Definition at line 195 of file mod_config.py.

◆ quantize_weight

qubee.configs.mod_config.ModConfig.quantize_weight = quantize_weight

Definition at line 196 of file mod_config.py.

◆ weight_scale_init

qubee.configs.mod_config.ModConfig.weight_scale_init = weight_scale_init

Definition at line 197 of file mod_config.py.

◆ downresol_mode

qubee.configs.mod_config.ModConfig.downresol_mode = downresol_mode

Definition at line 198 of file mod_config.py.

◆ post_processing

qubee.configs.mod_config.ModConfig.post_processing = post_processing

Definition at line 199 of file mod_config.py.

◆ box_conf_thres

qubee.configs.mod_config.ModConfig.box_conf_thres = box_conf_thres

Definition at line 200 of file mod_config.py.

◆ box_iou_thres

qubee.configs.mod_config.ModConfig.box_iou_thres = box_iou_thres

Definition at line 201 of file mod_config.py.

◆ loss_type

qubee.configs.mod_config.ModConfig.loss_type = loss_type

Definition at line 202 of file mod_config.py.

◆ use_outputs

qubee.configs.mod_config.ModConfig.use_outputs = use_outputs

Definition at line 203 of file mod_config.py.

◆ kl_temperature

qubee.configs.mod_config.ModConfig.kl_temperature = kl_temperature

Definition at line 204 of file mod_config.py.

◆ recon_prob

qubee.configs.mod_config.ModConfig.recon_prob = recon_prob

Definition at line 205 of file mod_config.py.

◆ recon_coeff

qubee.configs.mod_config.ModConfig.recon_coeff = recon_coeff

Definition at line 206 of file mod_config.py.

◆ lambda_0

qubee.configs.mod_config.ModConfig.lambda_0 = lambda_0

Definition at line 207 of file mod_config.py.

◆ lambda_1

qubee.configs.mod_config.ModConfig.lambda_1 = lambda_1

Definition at line 208 of file mod_config.py.

◆ lambda_2

qubee.configs.mod_config.ModConfig.lambda_2 = lambda_2

Definition at line 209 of file mod_config.py.

◆ anchors

qubee.configs.mod_config.ModConfig.anchors = anchors

Definition at line 210 of file mod_config.py.

◆ use_xyxy

qubee.configs.mod_config.ModConfig.use_xyxy = use_xyxy

Definition at line 211 of file mod_config.py.

◆ save_dir

qubee.configs.mod_config.ModConfig.save_dir = save_dir

Definition at line 212 of file mod_config.py.

◆ seed

qubee.configs.mod_config.ModConfig.seed = seed

Definition at line 213 of file mod_config.py.

◆ apply_layers

qubee.configs.mod_config.ModConfig.apply_layers = apply_layers

Definition at line 214 of file mod_config.py.

◆ exclude_layers

qubee.configs.mod_config.ModConfig.exclude_layers = exclude_layers

Definition at line 215 of file mod_config.py.

Mod Configuration

Mod Configuration#

Classes

Functions

Variables

Detailed Description

Function Documentation

◆ __init__()

◆ default_config()

◆ from_dict()

◆ to_dict()

◆ get_mod_config()

Variable Documentation

◆ optional

◆ DEFAULTS

◆ py_to_cpp_name

◆ min_max_check

◆ apply

◆ epochs

◆ warmup_epochs

◆ lr_act_scale

◆ lr_zeropoint

◆ lr_weight_scale

◆ lr_weight

◆ lr_bias

◆ lr_min_ratio

◆ batch_size

◆ q_drop

◆ quantize_weight

◆ weight_scale_init

◆ downresol_mode

◆ post_processing

◆ box_conf_thres

◆ box_iou_thres

◆ loss_type

◆ use_outputs

◆ kl_temperature

◆ recon_prob

◆ recon_coeff

◆ lambda_0

◆ lambda_1

◆ lambda_2

◆ anchors

◆ use_xyxy

◆ save_dir

◆ seed

◆ apply_layers

◆ exclude_layers

◆ init()