WIP: embeddings using PHI benchmarks

dvorka · Apr 6, 2024 · cc7453c · cc7453c
1 parent fdb6381
commit cc7453c
Show file tree

Hide file tree

Showing 29 changed files with 4,167 additions and 2,732 deletions.
diff --git a/Changelog b/Changelog
@@ -1,6 +1,13 @@
-2024-03-??  Martin Dvorak  <[email protected]>
-
-    * Released v2.1.0 - ...
+2024-04-??  Martin Dvorak  <[email protected]>
+
+    * Released v2.1.0 - Local LLMs support, Notes semantic search
+      and orphan library documents management.
+    * WIP: Feature: ollama integration brings privacy safe Wingman 
+      via ollama hosted LLMs.
+    * WIP: Feature: embeddings calculation by ollama hosted LLM models
+    * WIP: Enhancement: improved LLM providers configuration: OpenAI and ollama
+      (environment vs. configuration; LLM model selection; provider choice)
+    * WIP: Enhancement: library orphans
 
 2024-02-16  Martin Dvorak  <[email protected]>
 

diff --git a/app/resources/qt/translations/mindforger_cs.qm b/app/resources/qt/translations/mindforger_cs.qm
diff --git a/app/resources/qt/translations/mindforger_cs.ts b/app/resources/qt/translations/mindforger_cs.ts
diff --git a/app/resources/qt/translations/mindforger_en.qm b/app/resources/qt/translations/mindforger_en.qm
diff --git a/app/resources/qt/translations/mindforger_en.ts b/app/resources/qt/translations/mindforger_en.ts
diff --git a/app/resources/qt/translations/mindforger_nerd_cs.qm b/app/resources/qt/translations/mindforger_nerd_cs.qm
diff --git a/app/resources/qt/translations/mindforger_nerd_cs.ts b/app/resources/qt/translations/mindforger_nerd_cs.ts
diff --git a/app/resources/qt/translations/mindforger_nerd_en.qm b/app/resources/qt/translations/mindforger_nerd_en.qm
diff --git a/app/resources/qt/translations/mindforger_nerd_en.ts b/app/resources/qt/translations/mindforger_nerd_en.ts
diff --git a/app/resources/qt/translations/mindforger_zh_cn.qm b/app/resources/qt/translations/mindforger_zh_cn.qm
diff --git a/app/resources/qt/translations/mindforger_zh_cn.ts b/app/resources/qt/translations/mindforger_zh_cn.ts
diff --git a/app/src/qt/dialogs/configuration_dialog.cpp b/app/src/qt/dialogs/configuration_dialog.cpp
@@ -685,48 +685,30 @@ ConfigurationDialog::WingmanOpenAiTab::WingmanOpenAiTab(QWidget* parent, QComboB
       parentLlmProvidersCombo{parentLlmProvidersCombo},
       config(Configuration::getInstance())
 {
-    helpLabel = new QLabel(
-        tr(
-            "<html><a href='https://openai.com'>OpenAI</a> LLM provider configuration:\n"
-            "<ul>"
-            "<li>Generate new OpenAI API key at <a href='https://platform.openai.com/api-keys'>openai.com</a>.</li>"
-            "<li>a) either set the <b>%1</b> environment variable<br/>"
-                "with the API key<br/>"
-                "b) or paste the API key below to save it <font color='#ff0000'>unencrypted</font> to<br/>"
-                "<b>.mindforger.md</b> file in your home directory.</li>"
-            "</ul>"
-        ).arg(ENV_VAR_OPENAI_API_KEY));
-    BUG: if key in config > it overrides the env var > must be visible
-    hidden ONLY if env key is set & config key is empty
-    helpLabel->setVisible(!config.canWingmanOpenAiFromEnv());
     apiKeyLabel = new QLabel(tr("<br>API key:"));
-    apiKeyLabel->setVisible(helpLabel->isVisible());
     apiKeyEdit = new QLineEdit(this);
-    apiKeyEdit->setVisible(helpLabel->isVisible());
+    refreshLlmModelsButton = new QPushButton(tr("Refresh LLM models"), this);
+    refreshLlmModelsButton->setVisible(false);
     // enabled on valid config > add ollama to drop down > choose it in drop down
     setOpenAiButton = new QPushButton(tr("Set OpenAI as LLM Provider"), this);
     setOpenAiButton->setToolTip(
         tr("Add OpenAI to the dropdown with LLM providers (if not there) and set it for use with Wingman")
     );
-    setOpenAiButton->setVisible(helpLabel->isVisible());
     clearApiKeyButton = new QPushButton(tr("Clear API Key"), this);
-    clearApiKeyButton->setVisible(helpLabel->isVisible());
     // LLM model can be choose at any time when a valid configuration is available
     llmModelsLabel = new QLabel(tr("LLM model:"));
     llmModelsCombo = new QComboBox();
     llmModelsCombo->addItem(LLM_MODEL_NONE);
     llmModelsCombo->addItem(LLM_MODEL_GPT35_TURBO);
     llmModelsCombo->addItem(LLM_MODEL_GPT4);
 
-    configuredLabel = new QLabel(
-        tr("The OpenAI API key is configured using the environment variable."), this);
-    configuredLabel->setVisible(!helpLabel->isVisible());
+    // help text is created using initialized dialog fields
+    helpLabel = new QLabel(getHelpLabelText(), this);
 
     QVBoxLayout* llmProvidersLayout = new QVBoxLayout();
     llmProvidersLayout->addWidget(helpLabel);
     llmProvidersLayout->addWidget(apiKeyLabel);
     llmProvidersLayout->addWidget(apiKeyEdit);
-    llmProvidersLayout->addWidget(configuredLabel);
     llmProvidersLayout->addWidget(llmModelsLabel);
     llmProvidersLayout->addWidget(llmModelsCombo);
     QHBoxLayout* buttonsLayout = new QHBoxLayout{};
@@ -751,14 +733,56 @@ ConfigurationDialog::WingmanOpenAiTab::WingmanOpenAiTab(QWidget* parent, QComboB
 ConfigurationDialog::WingmanOpenAiTab::~WingmanOpenAiTab()
 {
     delete helpLabel;
-    delete configuredLabel;
     delete apiKeyLabel;
     delete apiKeyEdit;
+    delete refreshLlmModelsButton;
     delete clearApiKeyButton;
     delete llmModelsLabel;
     delete llmModelsCombo;
 }
 
+QString ConfigurationDialog::WingmanOpenAiTab::getHelpLabelText() const
+{
+    if(apiKeyEdit->text().size()==0) { 
+        if(config.canWingmanOpenAiFromEnv()) {
+            // key: NOT in config & IN env > ENV used
+            return tr(
+                "The OpenAI API key is configured using the environment<br/>"
+                "variable <b>%1</b>.<br/>"
+                "If you want to override it, then set the key below.<br/>"
+                "It will be saved <font color='#ff0000'>unencrypted</font> to "
+                "<b>.mindforger.md</b> configuration file in your home directory."
+            ).arg(ENV_VAR_OPENAI_API_KEY);
+        } else {
+            // key: NOT in config & NOT in env > NO key
+            return tr(
+                "<html><a href='https://openai.com'>OpenAI</a> LLM provider configuration:\n"
+                "<ul>"
+                "<li>Generate new OpenAI API key at <a href='https://platform.openai.com/api-keys'>openai.com</a>.</li>"
+                "<li>a) either set the <b>%1</b> environment variable<br/>"
+                    "with the API key<br/>"
+                    "b) or paste the API key below to save it <font color='#ff0000'>unencrypted</font> to<br/>"
+                    "<b>.mindforger.md</b> file in your home directory.</li>"
+                "</ul>"
+            ).arg(ENV_VAR_OPENAI_API_KEY);
+        }
+    } else {
+        if(config.canWingmanOpenAiFromEnv()) {
+            // key: IN config & IN env > CONFIG used
+            return tr(
+                "The OpenAI API key is configured both using the environment<br/>"
+                "variable <b>%1</b> and configuration.<br/>"
+                "Configuration overrides environment - key from below is used."
+            ).arg(ENV_VAR_OPENAI_API_KEY);
+        } else {
+            // key: IN config & NOT env > CONFIG used
+            return tr(
+                "The OpenAI API key is configured using the key below."
+            ).arg(ENV_VAR_OPENAI_API_KEY);
+        }
+    }
+}
+
 void refreshWingmanLlmProvidersComboBox(
     QComboBox* parentLlmProvidersCombo,
     Configuration& config
@@ -834,6 +858,8 @@ void ConfigurationDialog::WingmanOpenAiTab::refresh()
     apiKeyEdit->setText(QString::fromStdString(config.getWingmanOpenAiApiKey()));
     llmModelsCombo->setCurrentText(
             QString::fromStdString(config.getWingmanOpenAiLlm()));
+
+    helpLabel->setText(getHelpLabelText());
 }
 
 void ConfigurationDialog::WingmanOpenAiTab::save()
@@ -868,6 +894,7 @@ ConfigurationDialog::WingmanOllamaTab::WingmanOllamaTab(QWidget* parent, QComboB
     setOllamaButton->setToolTip(
         tr("Add ollama to the dropdown with LLM providers (if not there) and set it for use with Wingman")
     );
+    refreshLlmModelsButton = new QPushButton(tr("Refresh LLM models"), this);
     clearUrlButton = new QPushButton(tr("Clear URL"), this);
     llmModelsLabel = new QLabel(tr("LLM model:"));
     llmModelsCombo = new QComboBox();
@@ -880,6 +907,7 @@ ConfigurationDialog::WingmanOllamaTab::WingmanOllamaTab(QWidget* parent, QComboB
     llmProvidersLayout->addWidget(llmModelsLabel);
     llmProvidersLayout->addWidget(llmModelsCombo);
     QHBoxLayout* buttonsLayout = new QHBoxLayout{};
+    buttonsLayout->addWidget(refreshLlmModelsButton);
     buttonsLayout->addWidget(setOllamaButton);
     buttonsLayout->addWidget(clearUrlButton);
     llmProvidersLayout->addLayout(buttonsLayout);
@@ -891,6 +919,9 @@ ConfigurationDialog::WingmanOllamaTab::WingmanOllamaTab(QWidget* parent, QComboB
     layout->addStretch();
     setLayout(layout);
 
+    QObject::connect(
+        refreshLlmModelsButton, SIGNAL(clicked()),
+        this, SLOT(refreshLlmModelsSlot()));
     QObject::connect(
         setOllamaButton, SIGNAL(clicked()),
         this, SLOT(setOllamaSlot()));
@@ -904,11 +935,19 @@ ConfigurationDialog::WingmanOllamaTab::~WingmanOllamaTab()
     delete helpLabel;
     delete urlLabel;
     delete urlEdit;
+    delete refreshLlmModelsButton;
     delete clearUrlButton;
     delete llmModelsLabel;
     delete llmModelsCombo;
 }
 
+void ConfigurationDialog::WingmanOllamaTab::refreshLlmModelsSlot()
+{
+    MF_DEBUG("Signal SLOT: refresh LLM models" << endl);
+
+    // TODO: refresh LLM models - how to call mind cleanly?
+}
+
 void ConfigurationDialog::WingmanOllamaTab::setOllamaSlot()
 {
     MF_DEBUG("Signal SLOT: set ollama" << endl);

diff --git a/app/src/qt/dialogs/configuration_dialog.h b/app/src/qt/dialogs/configuration_dialog.h
@@ -90,9 +90,9 @@ class ConfigurationDialog::WingmanOpenAiTab : public QWidget
     Configuration& config;
 
     QLabel* helpLabel;
-    QLabel* configuredLabel;
     QLabel* apiKeyLabel;
     QLineEdit* apiKeyEdit;
+    QPushButton* refreshLlmModelsButton;
     QPushButton* setOpenAiButton;
     QPushButton* clearApiKeyButton;
     QLabel* llmModelsLabel;
@@ -105,6 +105,10 @@ class ConfigurationDialog::WingmanOpenAiTab : public QWidget
     void refresh();
     void save();
 
+private:
+
+QString getHelpLabelText() const;
+
 private slots:
     void setOpenAiSlot();
     void clearApiKeySlot();
@@ -125,6 +129,7 @@ class ConfigurationDialog::WingmanOllamaTab : public QWidget
     QLabel* helpLabel;
     QLabel* urlLabel;
     QLineEdit* urlEdit;
+    QPushButton* refreshLlmModelsButton;
     QPushButton* setOllamaButton;
     QPushButton* clearUrlButton;
     QLabel* llmModelsLabel;
@@ -138,6 +143,7 @@ class ConfigurationDialog::WingmanOllamaTab : public QWidget
     void save();
 
 private slots:
+    void refreshLlmModelsSlot();
     void setOllamaSlot();
     void clearUrlSlot();
 };

diff --git a/app/src/qt/main_menu_presenter.cpp b/app/src/qt/main_menu_presenter.cpp
@@ -45,6 +45,9 @@ MainMenuPresenter::MainMenuPresenter(MainWindowPresenter* mwp)
     QObject::connect(
         view->actionMindAutolink, SIGNAL(triggered()),
         mwp, SLOT(doActionMindToggleAutolink()));
+    QObject::connect(
+        view->actionMindSemanticSearch, SIGNAL(triggered()),
+        mwp, SLOT(doActionMindToggleSemanticSearch()));
     QObject::connect(
         view->actionMindLearnDirectory, SIGNAL(triggered()),
         mwp, SLOT(doActionMindLearnRepository()));

diff --git a/app/src/qt/main_menu_view.cpp b/app/src/qt/main_menu_view.cpp
@@ -109,6 +109,12 @@ MainMenuView::MainMenuView(MainWindowView& mainWindowView)
     actionMindAutolink->setVisible(false);
 #endif
 
+    // if Wingman is not configured and/or capable, then dialog with error message is shown (unable to detect it here)
+    actionMindSemanticSearch = new QAction(QIcon(":/menu-icons/find.svg"), tr("&Semantic Search"), mainWindow);
+    actionMindSemanticSearch->setCheckable(true);
+    actionMindSemanticSearch->setStatusTip(tr("Use Wingman LLM to search for similar Notes (associations) using text embeddings..."));
+    actionMindSemanticSearch->setShortcut(QKeySequence(Qt::CTRL+Qt::SHIFT+Qt::Key_S));    
+
     actionMindWingman = new QAction(QIcon(":/menu-icons/wingman-green.svg"), tr("&Wingman LLM"), mainWindow);
     actionMindWingman->setShortcut(QKeySequence(Qt::CTRL+Qt::Key_Slash));
     actionMindWingman->setStatusTip(tr("Open Wingman dialog..."));
@@ -205,6 +211,7 @@ MainMenuView::MainMenuView(MainWindowView& mainWindowView)
     menuMind->addSeparator();
     menuMind->addAction(actionMindThink);
     menuMind->addAction(actionMindAutolink);
+    menuMind->addAction(actionMindSemanticSearch);
     menuMind->addAction(actionMindWingman);
     menuMind->addAction(actionMindTool);
     menuMind->addAction(actionMindScope);

diff --git a/app/src/qt/main_menu_view.h b/app/src/qt/main_menu_view.h
@@ -77,6 +77,9 @@ class MainMenuView : public QObject
     QAction* actionMindRemember;
     QAction* actionMindThink;
     QAction* actionMindAutolink;
+#ifdef DO_MF_DEBUG
+    QAction* actionMindSemanticSearch;
+#endif
     QAction* actionMindWingman;
     QAction* actionMindTool;
     QAction* actionMindScope;

diff --git a/app/src/qt/main_window_presenter.cpp b/app/src/qt/main_window_presenter.cpp
@@ -666,6 +666,38 @@ void MainWindowPresenter::doActionMindToggleAutolink()
     }
 }
 
+void MainWindowPresenter::doActionMindToggleSemanticSearch()
+{
+    // TODO: to be implemented in analogous way as autolinking
+    if(config.isSemanticSearch()) {
+        config.setSemanticSearch(false);
+        statusBar->showInfo(tr("Semantic search disabled"));
+    } else {
+        // check whether possible
+        if(config.canWingmanOllama()) {
+            config.setSemanticSearch(true);
+            statusBar->showInfo(tr("Semantic search activated"));
+        } else {
+            config.setSemanticSearch(false);
+            statusBar->showError(tr("Semantic search cannot be activated - missing dependencies"));
+            QMessageBox::critical(
+                &view,
+                tr("Semantic Search"),
+                tr("Semantic search cannot be activated - ollama Wingman must be configured.")
+            );
+            return;
+        }
+    }
+    mdConfigRepresentation->save(config);
+
+    // activate semantic search
+    if(config.isSemanticSearch()) {
+        statusBar->showInfo(tr("Refresh semantic search index ~ text embeddings of all (modified) Notes..."));
+        mind->refreshEmbeddings();
+    }
+}
+
+
 void MainWindowPresenter::doActionNameDescFocusSwap()
 {
     if(orloj->isFacetActive(OrlojPresenterFacets::FACET_EDIT_OUTLINE_HEADER)) {

diff --git a/app/src/qt/main_window_presenter.h b/app/src/qt/main_window_presenter.h
@@ -195,6 +195,7 @@ public slots:
     void doActionMindSleep();
     void doActionMindToggleThink();
     void doActionMindToggleAutolink();
+    void doActionMindToggleSemanticSearch();
     void doActionMindLearnRepository();
     void doActionMindLearnFile();
     void doActionMindRelearn(QString path);

diff --git a/build/Makefile b/build/Makefile
@@ -147,9 +147,9 @@ build: clean-app ../app/mindforger ## build production MindForger application bi
 
 
 .PHONY: build-dev
-build-dev: clean-app ## build development MindForger application binary
+build-dev: clean-app ## build development MindForger application binary w/ debug info
 	@echo "Building DEV MindForger executable..."
-	cd .. && qmake -r mindforger.pro DEFINES+=DO_MF_DEBUG && make -j $(CPU_CORES) ; cd ..
+	cd .. && qmake -r mindforger.pro CONFIG+=debug DEFINES+=DO_MF_DEBUG && make -j $(CPU_CORES) ; cd ..
 ifeq ($(UNAME), Darwin)
 	@echo "\nIf build succeeded, then MindForger executable can be found in:\n  app/mindforger.app/Contents/MacOS/mindforger"
 	ls -al ../app/mindforger.app/Contents/MacOS/mindforger
@@ -204,6 +204,13 @@ else
 	# cd ../app && pwd && ./mindforger /home/dvorka/mf-devel/bug-copy-image
 endif
 
+run-gdb: build-dev ## run MindForger using gdb
+ifeq ($(UNAME), Darwin)
+	cd ../app/mindforger.app/Contents/MacOS && ./mindforger /Users/dvorka/mf-devel/mf-copy
+else
+	cd ../app && pwd && gdb --args ./mindforger /home/dvorka/mf-devel/library-trainer
+endif
+
 #
 # install
 #

diff --git a/lib/src/config/configuration.cpp b/lib/src/config/configuration.cpp
@@ -39,7 +39,7 @@ const string Configuration::DEFAULT_UI_HTML_CSS_THEME = string{UI_DEFAULT_HTML_C
 const string Configuration::DEFAULT_EDITOR_FONT= string{UI_DEFAULT_EDITOR_FONT};
 const string Configuration::DEFAULT_TIME_SCOPE = string{"0y0m0d0h0m"};
 const string Configuration::DEFAULT_WINGMAN_LLM_MODEL_OPENAI = string{LLM_MODEL_GPT35_TURBO};
-const string Configuration::DEFAULT_WINGMAN_LLM_MODEL_OLLAMA = string{LLM_MODEL_LLAMA2};
+const string Configuration::DEFAULT_WINGMAN_LLM_MODEL_OLLAMA = string{LLM_MODEL_PHI};
 
 Configuration::Configuration()
     : asyncMindThreshold{},
@@ -51,6 +51,7 @@ Configuration::Configuration()
       autolinking{DEFAULT_AUTOLINKING},
       autolinkingColonSplit{},
       autolinkingCaseInsensitive{},
+      semanticSearch{DEFAULT_SEMANTIC_SEARCH},
       wingmanProvider{DEFAULT_WINGMAN_LLM_PROVIDER},
       wingmanOpenAiApiKey{},
       wingmanOpenAiLlm{DEFAULT_WINGMAN_LLM_MODEL_OPENAI},
@@ -151,6 +152,7 @@ void Configuration::clear()
     autolinking = DEFAULT_AUTOLINKING;
     autolinkingColonSplit = DEFAULT_AUTOLINKING_COLON_SPLIT;
     autolinkingCaseInsensitive = DEFAULT_AUTOLINKING_CASE_INSENSITIVE;
+    semanticSearch = DEFAULT_SEMANTIC_SEARCH;
     wingmanProvider = DEFAULT_WINGMAN_LLM_PROVIDER;
     wingmanOpenAiApiKey.clear();
     wingmanOpenAiLlm = DEFAULT_WINGMAN_LLM_MODEL_OPENAI;

diff --git a/lib/src/config/configuration.h b/lib/src/config/configuration.h
@@ -57,6 +57,7 @@ constexpr const auto LLM_MODEL_GPT35_TURBO = "gpt-3.5-turbo";
 constexpr const auto LLM_MODEL_GPT4 = "gpt-4";
 // TODO ollama does NOT have to host llama2 > it should NOT be offered as default model
 constexpr const auto LLM_MODEL_LLAMA2 = "llama2";
+constexpr const auto LLM_MODEL_PHI = "phi";
 
 // const in constexpr makes value const
 constexpr const auto ENV_VAR_HOME = "HOME";
@@ -275,6 +276,8 @@ class Configuration {
     static constexpr const bool DEFAULT_AUTOLINKING_CASE_INSENSITIVE = true;
     static constexpr const bool DEFAULT_SAVE_READS_METADATA = true;
 
+    static constexpr const bool DEFAULT_SEMANTIC_SEARCH = false;
+
     static constexpr const bool UI_DEFAULT_NERD_TARGET_AUDIENCE = true;
     static const std::string DEFAULT_STARTUP_VIEW_NAME;
     static const std::string DEFAULT_UI_THEME_NAME;
@@ -343,6 +346,8 @@ class Configuration {
     bool autolinkingColonSplit;
     bool autolinkingCaseInsensitive;
 
+    bool semanticSearch;
+
     /*
     Wingman configuration, initialization and use:
 
@@ -544,6 +549,8 @@ class Configuration {
     void setAutolinkingColonSplit(bool autolinkingColonSplit) { this->autolinkingColonSplit=autolinkingColonSplit; }
     bool isAutolinkingCaseInsensitive() const { return autolinkingCaseInsensitive; }
     void setAutolinkingCaseInsensitive(bool autolinkingCaseInsensitive) { this->autolinkingCaseInsensitive=autolinkingCaseInsensitive; }
+    bool isSemanticSearch() const { return semanticSearch; }
+    void setSemanticSearch(bool semanticSearch) { this->semanticSearch=semanticSearch; }
     unsigned int getMd2HtmlOptions() const { return md2HtmlOptions; }
     AssociationAssessmentAlgorithm getAaAlgorithm() const { return aaAlgorithm; }
     void setAaAlgorithm(AssociationAssessmentAlgorithm aaa) { aaAlgorithm = aaa; }