This class allows loading only a subset of the weights of a large model. More...

#include <model-io.h>

Inheritance diagram for dismec::io::model::PartialModelLoader:

Classes
struct	SubModelRangeSpec

Public Types
enum	ESparseMode { DEFAULT , FORCE_SPARSE , FORCE_DENSE }

Public Member Functions
	PartialModelLoader (path meta_file, ESparseMode mode=DEFAULT)
	Create a new PartialModelLoader for the given metadata file. More...

const path &	meta_file_path () const
	The path to the metadata file. More...

std::shared_ptr< Model >	load_model (label_id_t label_begin, label_id_t label_end) const
	Loads part of the model. More...

std::shared_ptr< Model >	load_model (int index) const
	Loads the model from the weights file at the given index. More...

bool	validate () const
	Validates that all weight files exist. More...

long	num_weight_files () const
	Returns the number of availabel weight files. More...

SubModelRangeSpec	get_loading_range (label_id_t label_begin, label_id_t label_end) const

Public Member Functions inherited from dismec::io::model::PartialModelIO
long	num_labels () const noexcept
	Gets the total number of labels. More...

long	num_features () const noexcept
	Gets the total number of features. More...

Private Attributes
path	m_MetaFileName

ESparseMode	m_SparseMode

Additional Inherited Members
Protected Types inherited from dismec::io::model::PartialModelIO
using	weight_file_iter_t = std::vector< WeightFileEntry >::const_iterator

Protected Member Functions inherited from dismec::io::model::PartialModelIO
	PartialModelIO ()=default

	~PartialModelIO ()=default

void	read_metadata_file (const path &meta_file)

void	insert_sub_file (const WeightFileEntry &data)
	Inserts a new sub-file entry into the metadata object. More...

weight_file_iter_t	label_lower_bound (label_id_t pos) const
	Gets an iterator into the weight-file list that points to the first element whose starting label is larger than or equal to `pos`. More...

Protected Attributes inherited from dismec::io::model::PartialModelIO
long	m_TotalLabels = -1

long	m_NumFeatures = -1

std::vector< WeightFileEntry >	m_SubFiles

Detailed Description

This class allows loading only a subset of the weights of a large model.

In a setting with an extreme number of labels, it might not be possible to load the entire set of weights into the memory at once.

Definition at line 317 of file model-io.h.

Member Enumeration Documentation

◆ ESparseMode

enum dismec::io::model::PartialModelLoader::ESparseMode

Enumerator
DEFAULT
FORCE_SPARSE
FORCE_DENSE

Definition at line 319 of file model-io.h.

Constructor & Destructor Documentation

◆ PartialModelLoader()

PartialModelLoader::PartialModelLoader	(	path	meta_file,
		ESparseMode	mode = `DEFAULT`
	)

explicit

Create a new PartialModelLoader for the given metadata file.

This operation will only open and parse the metadata file, but no weights will be read in.

Parameters

meta_file	Path to the metadata file.
mode	Should the returned models store their weights in a sparse matrix, a dense matrix, or use the representation as given by the weights file (DEFAULT).

Definition at line 341 of file model-io.cpp.

References m_MetaFileName, and dismec::io::model::PartialModelIO::read_metadata_file().

Member Function Documentation

◆ get_loading_range()

auto PartialModelLoader::get_loading_range	(	label_id_t	label_begin,
		label_id_t	label_end
	)		const

Definition at line 346 of file model-io.cpp.

References dismec::io::model::WeightFileEntry::Count, dismec::io::model::WeightFileEntry::First, and dismec::io::model::NULL_FORMAT.

Referenced by load_model().

◆ load_model() [1/2]

std::shared_ptr< Model > PartialModelLoader::load_model ( int index ) const

Loads the model from the weights file at the given index.

Parameters

index The index of the sub-model to be loaded.

Definition at line 423 of file model-io.cpp.

References dismec::io::model::WeightFileEntry::Count, dismec::io::model::WeightFileEntry::FileName, dismec::io::model::WeightFileEntry::First, dismec::io::model::WeightFileEntry::Format, m_SparseMode, dismec::io::model::PartialModelIO::m_SubFiles, anonymous_namespace{model-io.cpp}::make_model(), meta_file_path(), dismec::io::model::PartialModelIO::num_features(), dismec::io::model::PartialModelIO::num_labels(), anonymous_namespace{model-io.cpp}::read_weights_dispatch(), THROW_ERROR, and anonymous_namespace{model-io.cpp}::use_sparse_weights().

◆ load_model() [2/2]

std::shared_ptr< Model > PartialModelLoader::load_model	(	label_id_t	label_begin,
		label_id_t	label_end
	)		const

Loads part of the model.

If the desired interval [label_begin, label_end) does not match directly with the weight files, a model that encompasses more weights will be used. Each weights file which has some overlap with the interval will be loaded entirely.

Parameters

label_begin	First label whose weight is guaranteed to be part of the returned model.
label_end	Element id after the last weight vector that is included in the returned model.

Returns: A model that contains weight at least for the interval [label_begin, label_end).

Definition at line 373 of file model-io.cpp.

References get_loading_range(), m_MetaFileName, dismec::io::model::PartialModelIO::m_NumFeatures, dismec::io::model::PartialModelIO::m_TotalLabels, anonymous_namespace{model-io.cpp}::read_weights_dispatch(), and THROW_ERROR.

Referenced by dismec::io::model::load_model(), and main().

◆ meta_file_path()

const path& dismec::io::model::PartialModelLoader::meta_file_path ( ) const

inline

The path to the metadata file.

Definition at line 338 of file model-io.h.

References m_MetaFileName.

Referenced by load_model(), and validate().

◆ num_weight_files()

long PartialModelLoader::num_weight_files ( ) const

Returns the number of availabel weight files.

Definition at line 395 of file model-io.cpp.

References dismec::io::model::PartialModelIO::m_SubFiles.

Referenced by main().

◆ validate()

bool PartialModelLoader::validate ( ) const

Validates that all weight files exist.

Definition at line 441 of file model-io.cpp.

References dismec::io::model::PartialModelIO::m_SubFiles, and meta_file_path().

Referenced by main().

Member Data Documentation

◆ m_MetaFileName

path dismec::io::model::PartialModelLoader::m_MetaFileName

private

Definition at line 374 of file model-io.h.

Referenced by load_model(), meta_file_path(), and PartialModelLoader().

◆ m_SparseMode

ESparseMode dismec::io::model::PartialModelLoader::m_SparseMode

private

Definition at line 375 of file model-io.h.

Referenced by load_model().

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

src/io/model-io.h
src/io/model-io.cpp

Classes

Public Types

Public Member Functions

Private Attributes

Additional Inherited Members

Detailed Description

Member Enumeration Documentation

◆ ESparseMode

Constructor & Destructor Documentation

◆ PartialModelLoader()

Member Function Documentation

◆ get_loading_range()

◆ load_model() [1/2]

◆ load_model() [2/2]

◆ meta_file_path()

◆ num_weight_files()

◆ validate()

Member Data Documentation

◆ m_MetaFileName

◆ m_SparseMode