Loading...
Searching...
No Matches
opennn::WordPieceTokenizer Class Reference

#include <opennn/network/operators/tokenizer_operator.h>

Inheritance diagram for opennn::WordPieceTokenizer:
opennn::TokenizerOperator opennn::Operator

Public Member Functions

 WordPieceTokenizer ()=default
 
 WordPieceTokenizer (const vector< string > &vocabulary)
 
void load_vocabulary (const filesystem::path &vocabulary_file)
 
void set_vocabulary (const vector< string > &) override
 
vector< string > tokenize (string_view text) const override
 
vector< Index > encode (string_view text) const override
 
void build_vocabulary (const vector< vector< string > > &, Index, Index) override
 
void set_lower_case (bool value) noexcept
 
unique_ptr< TokenizerOperator > clone () const override
 
string_view get_kind () const override
 
uint64_t fingerprint () const override
 
void to_JSON (JsonWriter &) const override
 
void from_JSON (const Json *) override
 
- Public Member Functions inherited from opennn::TokenizerOperator
const vector< string > & get_vocabulary () const noexcept
 
const vector< string > & get_reserved_tokens () const noexcept
 
const VocabularyMap & get_vocabulary_map () const noexcept
 
Index get_vocabulary_size () const noexcept
 
Index token_to_id (string_view token) const
 
const string & id_to_token (Index id) const
 
vector< Index > encode_sequence (const vector< string > &tokens, Index sequence_length) const
 
vector< Index > encode_sequence (string_view text, Index sequence_length) const
 
virtual string decode (const vector< Index > &ids) const
 
virtual string decode_token (Index id) const
 
virtual bool supports_incremental_decode () const noexcept
 
Index get_unk_id () const noexcept
 
- Public Member Functions inherited from opennn::Operator
virtual ~Operator ()=default
 
virtual vector< TensorSpec > parameter_specs () const
 
virtual vector< TensorSpec > state_specs () const
 
virtual vector< SlotQuantization > parameter_quantization () const
 
virtual vector< ParameterSlot > parameter_slots ()
 
virtual void link_parameters (span< const TensorView > views)
 
virtual void link_gradients (span< const TensorView > views)
 
virtual void link_states (span< const TensorView >)
 
virtual void link_parameter_scales (span< const TensorView >)
 
bool link_slots (span< const TensorView > views, TensorView *ParameterSlot::*member)
 
virtual void initialize_states ()
 
virtual void set_weights_dtype (Type new_weights_dtype)
 
virtual void set_parameters_random ()
 
virtual void set_parameters_glorot ()
 
virtual void set_parameters_pytorch ()
 
virtual void forward_propagate (ForwardPropagation &, size_t, ForwardPropagationMode)
 
virtual void back_propagate (ForwardPropagation &, BackPropagation &, size_t) const
 
virtual void load_state_from_JSON (const Json *)
 
TensorView & get_input (ForwardPropagation &forward_propagation, size_t layer, size_t slot_index=0) const noexcept
 
vector< TensorView > & get_inputs (ForwardPropagation &forward_propagation, size_t layer) const noexcept
 
TensorView & get_output (ForwardPropagation &forward_propagation, size_t layer, size_t slot_index=0) const noexcept
 
TensorView & get_output_delta (BackPropagation &back_propagation, size_t layer, size_t slot_index=0) const noexcept
 
TensorView & get_input_delta (BackPropagation &back_propagation, size_t layer, size_t slot_index=0) const noexcept
 

Additional Inherited Members

- Public Types inherited from opennn::TokenizerOperator
using VocabularyMap = StringMap<Index>
 
- Public Attributes inherited from opennn::Operator
Type compute_dtype = Type::FP32
 
Type weights_dtype = Type::FP32
 
vector< size_t > input_slots = {0}
 
vector< size_t > output_slots = {1}
 
vector< size_t > input_delta_slots = {1}
 
vector< size_t > output_delta_slots = {0}
 
- Static Public Attributes inherited from opennn::TokenizerOperator
static constexpr Index UNK_INDEX = 1
 
static constexpr Index START_INDEX = 2
 
static constexpr Index END_INDEX = 3
 
- Protected Member Functions inherited from opennn::TokenizerOperator
void rebuild_map ()
 
vector< Index > frame_sequence (span< const Index >, Index sequence_length) const
 
- Protected Attributes inherited from opennn::TokenizerOperator
vector< string > reserved_tokens
 
Index unk_id = 0
 
Index start_id = -1
 
Index end_id = -1
 
vector< string > vocabulary
 
VocabularyMap vocabulary_map
 

Constructor & Destructor Documentation

◆ WordPieceTokenizer() [1/2]

opennn::WordPieceTokenizer::WordPieceTokenizer ( )
default

◆ WordPieceTokenizer() [2/2]

opennn::WordPieceTokenizer::WordPieceTokenizer ( const vector< string > & vocabulary)
explicit

Member Function Documentation

◆ build_vocabulary()

void opennn::WordPieceTokenizer::build_vocabulary ( const vector< vector< string > > & ,
Index ,
Index  )
inlineoverridevirtual

Reimplemented from opennn::TokenizerOperator.

◆ clone()

unique_ptr< TokenizerOperator > opennn::WordPieceTokenizer::clone ( ) const
inlineoverridevirtual

◆ encode()

vector< Index > opennn::WordPieceTokenizer::encode ( string_view text) const
overridevirtual

Reimplemented from opennn::TokenizerOperator.

◆ fingerprint()

uint64_t opennn::WordPieceTokenizer::fingerprint ( ) const
overridevirtual

Reimplemented from opennn::TokenizerOperator.

◆ from_JSON()

void opennn::WordPieceTokenizer::from_JSON ( const Json * )
overridevirtual

Reimplemented from opennn::TokenizerOperator.

◆ get_kind()

string_view opennn::WordPieceTokenizer::get_kind ( ) const
inlineoverridevirtual

◆ load_vocabulary()

void opennn::WordPieceTokenizer::load_vocabulary ( const filesystem::path & vocabulary_file)

◆ set_lower_case()

void opennn::WordPieceTokenizer::set_lower_case ( bool value)
inlinenoexcept

◆ set_vocabulary()

void opennn::WordPieceTokenizer::set_vocabulary ( const vector< string > & )
overridevirtual

Reimplemented from opennn::TokenizerOperator.

◆ to_JSON()

void opennn::WordPieceTokenizer::to_JSON ( JsonWriter & ) const
overridevirtual

Reimplemented from opennn::TokenizerOperator.

◆ tokenize()

vector< string > opennn::WordPieceTokenizer::tokenize ( string_view text) const
overridevirtual