Digikam::SearchLanguageBackend Class Reference

Inheritance diagram for Digikam::SearchLanguageBackend:

Public Slots
virtual void	slotCancel ()
virtual void	slotRunInference (const QString &prompt)=0
	Run inference asynchronously.

Signals
void	signalInferenceCancelled ()
void	signalInferenceError (const QString &errorMessage)
void	signalInferenceProgress (int step)
void	signalModelLoaded (bool success)
void	signalModelLoadProgress (int percent)
void	signalRawOutputReady (const QString &output)

Public Member Functions
	SearchLanguageBackend (QObject *const parent=nullptr)
virtual QString	backendName () const =0
	Human-readable backend identifier ("llama.cpp", "onnx", "mock").
virtual bool	isModelLoaded () const =0
virtual bool	loadModel (const QString &modelPath)=0
	Load the model from a local path (resolved by SearchNlModelManager).
virtual QString	modelPath () const =0
virtual void	unloadModel ()=0

Member Function Documentation

backendName()

virtual QString Digikam::SearchLanguageBackend::backendName ( ) const

pure virtual

Implemented in Digikam::SearchLlamaBackend, and Digikam::SearchMockBackend.

loadModel()

virtual bool Digikam::SearchLanguageBackend::loadModel ( const QString & modelPath )

pure virtual

Implementations should perform heavy loading off the GUI thread and report completion through signalModelLoaded(). Returns false immediately if the path is invalid.

Implemented in Digikam::SearchLlamaBackend, and Digikam::SearchMockBackend.

slotRunInference

virtual void Digikam::SearchLanguageBackend::slotRunInference ( const QString & prompt )

pure virtualslot

Result is delivered via signalRawOutputReady() or signalInferenceError(). Implementations must be safe to call from the GUI thread and must not block it.