qbcompiler.configs.models.CompileConfig Class Reference

qbcompiler.configs.models.CompileConfig Class Reference#

Mobilint SDK qb Compiler: qbcompiler.configs.models.CompileConfig Class Reference

Mobilint SDK qb Compiler v1.2

MCS002-KR

Unified compilation configuration for Mobilint MXQ compilation. More...

Inheritance diagram for qbcompiler.configs.models.CompileConfig:

Public Member Functions
"CompileConfig"	with_uint8_input (self, **kwargs)
	Return a copy with uint8_input settings enabled.
"CompileConfig"	with_preprocessing (self, **kwargs)
	Return a copy with preprocessing settings enabled.
"CompileConfig"	with_optq (self, **kwargs)
	Return a copy with optq settings enabled.
"CompileConfig"	with_mod (self, **kwargs)
	Return a copy with mod settings enabled.
"CompileConfig"	with_llm (self, **kwargs)
	Return a copy with llm settings enabled.
"CompileConfig"	with_search_weight_scale (self, **kwargs)
	Return a copy with search_weight_scale settings enabled.
"CompileConfig"	with_load_scale (self, **kwargs)
	Return a copy with load_scale settings enabled.
"CompileConfig"	with_save_sample (self, **kwargs)
	Return a copy with save_sample settings enabled.
"CompileConfig"	from_file (cls, Union[str, Path] path)
	Load config from YAML or JSON file.
"CompileConfig"	from_preset (cls, str name)
	Load config from a preset.
None	to_file (self, Union[str, Path] path)
	Save config to YAML or JSON file.

Static Public Attributes
	model_config
List	model_paths
List	calib_data_path
List	save_paths
bool	use_random_calib
Optional	save_msgpack_name
str	inference_scheme
bool	cpu_offload
bool	force_npu_input_reposition
bool	force_npu_output_reposition
int	optimize_option
int	buffer_mode
Any	input_shape_dict
str	device = Field(default="gpu", description="Device for computation")
str	dtype = Field(default="float", description="Data type for computation")
bool	debug = Field(default=False, description="Enable debug mode")
bool	trace = Field(default=False, description="Enable trace mode")
int	image_channels
str	config_version
List	split_blocks
int	split_parts
Uint8InputConfig	uint8_input
PreprocessingConfig	preprocessing = Field(default_factory=PreprocessingConfig)
ResourceManagementConfig	resource_management
CalibrationConfig	calibration = Field(default_factory=CalibrationConfig)
BitConfig	bit = Field(default_factory=BitConfig)
OptqConfig	optq = Field(default_factory=OptqConfig)
ModConfig	mod = Field(default_factory=ModConfig)
LlmConfig	llm = Field(default_factory=LlmConfig)
MoeConfig	moe = Field(default_factory=MoeConfig)
EquivalentTransformationConfig	equivalent_transformation
SearchWeightScaleConfig	search_weight_scale
LoadScaleConfig	load_scale
RuntimeOptions	runtime_options
SaveSampleConfig	save_sample

Static Protected Member Functions
dict	_flatten_grouped_json (dict data)
	Flatten grouped JSON keys (e.g.
dict	_group_to_json (dict data)
	Group flat keys back into nested JSON structure (inverse of _flatten_grouped_json).

Detailed Description

Unified compilation configuration for Mobilint MXQ compilation.

Definition at line 1580 of file models.py.

Member Function Documentation

◆ with_uint8_input()

"CompileConfig" qbcompiler.configs.models.CompileConfig.with_uint8_input	(		self,
		**	kwargs )

Return a copy with uint8_input settings enabled.

Definition at line 1689 of file models.py.

◆ with_preprocessing()

"CompileConfig" qbcompiler.configs.models.CompileConfig.with_preprocessing	(		self,
		**	kwargs )

Return a copy with preprocessing settings enabled.

Definition at line 1695 of file models.py.

◆ with_optq()

"CompileConfig" qbcompiler.configs.models.CompileConfig.with_optq	(		self,
		**	kwargs )

Return a copy with optq settings enabled.

Definition at line 1701 of file models.py.

◆ with_mod()

"CompileConfig" qbcompiler.configs.models.CompileConfig.with_mod	(		self,
		**	kwargs )

Return a copy with mod settings enabled.

Definition at line 1707 of file models.py.

◆ with_llm()

"CompileConfig" qbcompiler.configs.models.CompileConfig.with_llm	(		self,
		**	kwargs )

Return a copy with llm settings enabled.

Definition at line 1713 of file models.py.

◆ with_search_weight_scale()

"CompileConfig" qbcompiler.configs.models.CompileConfig.with_search_weight_scale	(		self,
		**	kwargs )

Return a copy with search_weight_scale settings enabled.

Definition at line 1719 of file models.py.

◆ with_load_scale()

"CompileConfig" qbcompiler.configs.models.CompileConfig.with_load_scale	(		self,
		**	kwargs )

Return a copy with load_scale settings enabled.

Definition at line 1725 of file models.py.

◆ with_save_sample()

"CompileConfig" qbcompiler.configs.models.CompileConfig.with_save_sample	(		self,
		**	kwargs )

Return a copy with save_sample settings enabled.

Definition at line 1731 of file models.py.

◆ from_file()

"CompileConfig" qbcompiler.configs.models.CompileConfig.from_file	(		cls,
		Union[str, Path]	path )

Load config from YAML or JSON file.

Definition at line 1738 of file models.py.

◆ _flatten_grouped_json()

dict qbcompiler.configs.models.CompileConfig._flatten_grouped_json ( dict data )

staticprotected

Flatten grouped JSON keys (e.g.

quantization.calibration) to flat structure.

Definition at line 1750 of file models.py.

◆ from_preset()

"CompileConfig" qbcompiler.configs.models.CompileConfig.from_preset	(		cls,
		str	name )

Load config from a preset.

Definition at line 1774 of file models.py.

◆ _group_to_json()

dict qbcompiler.configs.models.CompileConfig._group_to_json ( dict data )

staticprotected

Group flat keys back into nested JSON structure (inverse of _flatten_grouped_json).

Definition at line 1781 of file models.py.

◆ to_file()

None qbcompiler.configs.models.CompileConfig.to_file	(		self,
		Union[str, Path]	path )

Save config to YAML or JSON file.

Definition at line 1810 of file models.py.

Member Data Documentation

◆ model_config

qbcompiler.configs.models.CompileConfig.model_config

static

Initial value:

=  ConfigDict(
        populate_by_name=True,
        extra="forbid",
    )

Definition at line 1583 of file models.py.

◆ model_paths

List qbcompiler.configs.models.CompileConfig.model_paths

static

Initial value:

=  Field(
        default=[], alias="modelPaths", description="Paths to model files"
    )

Definition at line 1588 of file models.py.

◆ calib_data_path

List qbcompiler.configs.models.CompileConfig.calib_data_path

static

Initial value:

=  Field(
        default=[], alias="calibDataPaths", description="Paths to calibration datasets"
    )

Definition at line 1591 of file models.py.

◆ save_paths

List qbcompiler.configs.models.CompileConfig.save_paths

static

Initial value:

=  Field(
        default=["./tmp.mxq"],
        alias="savePaths",
        description="Output MXQ filename/paths",
    )

Definition at line 1594 of file models.py.

◆ use_random_calib

bool qbcompiler.configs.models.CompileConfig.use_random_calib

static

Initial value:

=  Field(
        default=False, alias="useRandomCalib", description="Use random calibration"
    )

Definition at line 1599 of file models.py.

◆ save_msgpack_name

Optional qbcompiler.configs.models.CompileConfig.save_msgpack_name

static

Initial value:

=  Field(
        default=None,
        alias="saveMsgpackName",
        description="Name of the msgpack file to save",
    )

Definition at line 1602 of file models.py.

◆ inference_scheme

str qbcompiler.configs.models.CompileConfig.inference_scheme

static

Initial value:

=  Field(
        default="single", alias="inferenceScheme", description="NPU inference scheme"
    )

Definition at line 1607 of file models.py.

◆ cpu_offload

bool qbcompiler.configs.models.CompileConfig.cpu_offload

static

Initial value:

=  Field(
        default=False,
        alias="cpuOffload",
        description="Enable CPU offload for unsupported operators",
    )

Definition at line 1610 of file models.py.

◆ force_npu_input_reposition

bool qbcompiler.configs.models.CompileConfig.force_npu_input_reposition

static

Initial value:

=  Field(
        default=False,
        alias="forceNpuInputReposition",
        description="Force input reposition operations to run on NPU instead of CPU",
    )

Definition at line 1615 of file models.py.

◆ force_npu_output_reposition

bool qbcompiler.configs.models.CompileConfig.force_npu_output_reposition

static

Initial value:

=  Field(
        default=False,
        alias="forceNpuOutputReposition",
        description="Force output reposition operations to run on NPU instead of CPU",
    )

Definition at line 1620 of file models.py.

◆ optimize_option

int qbcompiler.configs.models.CompileConfig.optimize_option

static

Initial value:

=  Field(
        default=1,
        alias="optimizeOption",
        description="Compiler optimization selector",
        ge=0,
    )

Definition at line 1625 of file models.py.

◆ buffer_mode

int qbcompiler.configs.models.CompileConfig.buffer_mode

static

Initial value:

=  Field(
        default=1, alias="bufferMode", description="Buffer serialization mode"
    )

Definition at line 1631 of file models.py.

◆ input_shape_dict

Any qbcompiler.configs.models.CompileConfig.input_shape_dict

static

Initial value:

=  Field(
        default={}, alias="inputShapeDict", description="Dictionary of input shapes"
    )

Definition at line 1634 of file models.py.

◆ device

str qbcompiler.configs.models.CompileConfig.device = Field(default="gpu", description="Device for computation")

static

Definition at line 1637 of file models.py.

◆ dtype

str qbcompiler.configs.models.CompileConfig.dtype = Field(default="float", description="Data type for computation")

static

Definition at line 1638 of file models.py.

◆ debug

bool qbcompiler.configs.models.CompileConfig.debug = Field(default=False, description="Enable debug mode")

static

Definition at line 1639 of file models.py.

◆ trace

bool qbcompiler.configs.models.CompileConfig.trace = Field(default=False, description="Enable trace mode")

static

Definition at line 1640 of file models.py.

◆ image_channels

int qbcompiler.configs.models.CompileConfig.image_channels

static

Initial value:

=  Field(
        default=0,
        alias="imageChannels",
        description="Number of image channels (0 for auto-detect)",
    )

Definition at line 1641 of file models.py.

◆ config_version

str qbcompiler.configs.models.CompileConfig.config_version

static

Initial value:

=  Field(
        default="1.0.0", alias="configVersion", description="Config schema version"
    )

Definition at line 1646 of file models.py.

◆ split_blocks

List qbcompiler.configs.models.CompileConfig.split_blocks

static

Initial value:

=  Field(
        default=[],
        alias="splitBlocks",
        description="Multi-MXQ split points by transformer block index",
    )

Definition at line 1649 of file models.py.

◆ split_parts

int qbcompiler.configs.models.CompileConfig.split_parts

static

Initial value:

=  Field(
        default=0,
        alias="splitParts",
        description="Evenly split transformer blocks into N MXQ parts",
    )

Definition at line 1654 of file models.py.

◆ uint8_input

Uint8InputConfig qbcompiler.configs.models.CompileConfig.uint8_input

static

Initial value:

=  Field(
        default_factory=Uint8InputConfig, alias="uint8Input"
    )

Definition at line 1660 of file models.py.

◆ preprocessing

PreprocessingConfig qbcompiler.configs.models.CompileConfig.preprocessing = Field(default_factory=PreprocessingConfig)

static

Definition at line 1663 of file models.py.

◆ resource_management

ResourceManagementConfig qbcompiler.configs.models.CompileConfig.resource_management

static

Initial value:

=  Field(
        default_factory=ResourceManagementConfig, alias="resourceManagement"
    )

Definition at line 1664 of file models.py.

◆ calibration

CalibrationConfig qbcompiler.configs.models.CompileConfig.calibration = Field(default_factory=CalibrationConfig)

static

Definition at line 1667 of file models.py.

◆ bit

BitConfig qbcompiler.configs.models.CompileConfig.bit = Field(default_factory=BitConfig)

static

Definition at line 1668 of file models.py.

◆ optq

OptqConfig qbcompiler.configs.models.CompileConfig.optq = Field(default_factory=OptqConfig)

static

Definition at line 1669 of file models.py.

◆ mod

ModConfig qbcompiler.configs.models.CompileConfig.mod = Field(default_factory=ModConfig)

static

Definition at line 1670 of file models.py.

◆ llm

LlmConfig qbcompiler.configs.models.CompileConfig.llm = Field(default_factory=LlmConfig)

static

Definition at line 1671 of file models.py.

◆ moe

MoeConfig qbcompiler.configs.models.CompileConfig.moe = Field(default_factory=MoeConfig)

static

Definition at line 1672 of file models.py.

◆ equivalent_transformation

EquivalentTransformationConfig qbcompiler.configs.models.CompileConfig.equivalent_transformation

static

Initial value:

=  Field(
        default_factory=EquivalentTransformationConfig, alias="equivalentTransformation"
    )

Definition at line 1673 of file models.py.

◆ search_weight_scale

SearchWeightScaleConfig qbcompiler.configs.models.CompileConfig.search_weight_scale

static

Initial value:

=  Field(
        default_factory=SearchWeightScaleConfig, alias="searchWeightScale"
    )

Definition at line 1676 of file models.py.

◆ load_scale

LoadScaleConfig qbcompiler.configs.models.CompileConfig.load_scale

static

Initial value:

=  Field(
        default_factory=LoadScaleConfig, alias="loadScale"
    )

Definition at line 1679 of file models.py.

◆ runtime_options

RuntimeOptions qbcompiler.configs.models.CompileConfig.runtime_options

static

Initial value:

=  Field(
        default_factory=RuntimeOptions, alias="runtimeOptions"
    )

Definition at line 1682 of file models.py.

◆ save_sample

SaveSampleConfig qbcompiler.configs.models.CompileConfig.save_sample

static

Initial value:

=  Field(
        default_factory=SaveSampleConfig, alias="saveSample"
    )

Definition at line 1685 of file models.py.

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

models.py

qbcompiler.configs.models.CompileConfig Class Reference

qbcompiler.configs.models.CompileConfig Class Reference#

Public Member Functions

Static Public Attributes

Static Protected Member Functions

Detailed Description

Member Function Documentation

◆ with_uint8_input()

◆ with_preprocessing()

◆ with_optq()

◆ with_mod()

◆ with_llm()

◆ with_search_weight_scale()

◆ with_load_scale()

◆ with_save_sample()

◆ from_file()

◆ _flatten_grouped_json()

◆ from_preset()

◆ _group_to_json()

◆ to_file()

Member Data Documentation

◆ model_config

◆ model_paths

◆ calib_data_path

◆ save_paths

◆ use_random_calib

◆ save_msgpack_name

◆ inference_scheme

◆ cpu_offload

◆ force_npu_input_reposition

◆ force_npu_output_reposition

◆ optimize_option

◆ buffer_mode

◆ input_shape_dict

◆ device

◆ dtype

◆ debug

◆ trace

◆ image_channels

◆ config_version

◆ split_blocks

◆ split_parts

◆ uint8_input

◆ preprocessing

◆ resource_management

◆ calibration

◆ bit

◆ optq

◆ mod

◆ llm

◆ moe

◆ equivalent_transformation

◆ search_weight_scale

◆ load_scale

◆ runtime_options

◆ save_sample