chat: generate follow-up questions after response (#2634)

* user can configure the prompt and when they appear
* also make the name generation prompt configurable

Signed-off-by: Adam Treat <treat.adam@gmail.com>
Signed-off-by: Jared Van Bortel <jared@nomic.ai>
Co-authored-by: Jared Van Bortel <jared@nomic.ai>

gpt4all-chat/CMakeLists.txt

@@ -210,11 +210,13 @@ qt_add_qml_module(chat
       icons/network.svg
       icons/nomic_logo.svg
       icons/notes.svg
+      icons/plus.svg
       icons/recycle.svg
       icons/regenerate.svg
       icons/search.svg
       icons/send_message.svg
       icons/settings.svg
+      icons/stack.svg
       icons/stop_generating.svg
       icons/thumbs_down.svg
       icons/thumbs_up.svg

gpt4all-chat/chat.cpp

@@ -58,11 +58,13 @@ void Chat::connectLLM()
     connect(m_llmodel, &ChatLLM::modelLoadingPercentageChanged, this, &Chat::handleModelLoadingPercentageChanged, Qt::QueuedConnection);
     connect(m_llmodel, &ChatLLM::responseChanged, this, &Chat::handleResponseChanged, Qt::QueuedConnection);
     connect(m_llmodel, &ChatLLM::promptProcessing, this, &Chat::promptProcessing, Qt::QueuedConnection);
+    connect(m_llmodel, &ChatLLM::generatingQuestions, this, &Chat::generatingQuestions, Qt::QueuedConnection);
     connect(m_llmodel, &ChatLLM::responseStopped, this, &Chat::responseStopped, Qt::QueuedConnection);
     connect(m_llmodel, &ChatLLM::modelLoadingError, this, &Chat::handleModelLoadingError, Qt::QueuedConnection);
     connect(m_llmodel, &ChatLLM::modelLoadingWarning, this, &Chat::modelLoadingWarning, Qt::QueuedConnection);
     connect(m_llmodel, &ChatLLM::recalcChanged, this, &Chat::handleRecalculating, Qt::QueuedConnection);
     connect(m_llmodel, &ChatLLM::generatedNameChanged, this, &Chat::generatedNameChanged, Qt::QueuedConnection);
+    connect(m_llmodel, &ChatLLM::generatedQuestionFinished, this, &Chat::generatedQuestionFinished, Qt::QueuedConnection);
     connect(m_llmodel, &ChatLLM::reportSpeed, this, &Chat::handleTokenSpeedChanged, Qt::QueuedConnection);
     connect(m_llmodel, &ChatLLM::loadedModelInfoChanged, this, &Chat::loadedModelInfoChanged, Qt::QueuedConnection);
     connect(m_llmodel, &ChatLLM::databaseResultsChanged, this, &Chat::handleDatabaseResultsChanged, Qt::QueuedConnection);
@@ -113,6 +115,8 @@ void Chat::resetResponseState()
     if (m_responseInProgress && m_responseState == Chat::LocalDocsRetrieval)
         return;
 
+    m_generatedQuestions = QList<QString>();
+    emit generatedQuestionsChanged();
     m_tokenSpeed = QString();
     emit tokenSpeedChanged();
     m_responseInProgress = true;
@@ -189,6 +193,12 @@ void Chat::promptProcessing()
      emit responseStateChanged();
 }
 
+void Chat::generatingQuestions()
+{
+    m_responseState = Chat::GeneratingQuestions;
+    emit responseStateChanged();
+}
+
 void Chat::responseStopped(qint64 promptResponseMs)
 {
     m_tokenSpeed = QString();
@@ -304,6 +314,12 @@ void Chat::generatedNameChanged(const QString &name)
     emit nameChanged();
 }
 
+void Chat::generatedQuestionFinished(const QString &question)
+{
+    m_generatedQuestions << question;
+    emit generatedQuestionsChanged();
+}
+
 void Chat::handleRecalculating()
 {
     Network::globalInstance()->trackChatEvent("recalc_context", { {"length", m_chatModel->count()} });

gpt4all-chat/chat.h

@@ -39,6 +39,7 @@ class Chat : public QObject
     Q_PROPERTY(LocalDocsCollectionsModel *collectionModel READ collectionModel NOTIFY collectionModelChanged)
     // 0=no, 1=waiting, 2=working
     Q_PROPERTY(int trySwitchContextInProgress READ trySwitchContextInProgress NOTIFY trySwitchContextInProgressChanged)
+    Q_PROPERTY(QList<QString> generatedQuestions READ generatedQuestions NOTIFY generatedQuestionsChanged)
     QML_ELEMENT
     QML_UNCREATABLE("Only creatable from c++!")
 
@@ -48,6 +49,7 @@ public:
         LocalDocsRetrieval,
         LocalDocsProcessing,
         PromptProcessing,
+        GeneratingQuestions,
         ResponseGeneration
     };
     Q_ENUM(ResponseState)
@@ -119,6 +121,8 @@ public:
 
     int trySwitchContextInProgress() const { return m_trySwitchContextInProgress; }
 
+    QList<QString> generatedQuestions() const { return m_generatedQuestions; }
+
 public Q_SLOTS:
     void serverNewPromptResponsePair(const QString &prompt);
 
@@ -153,13 +157,16 @@ Q_SIGNALS:
     void collectionModelChanged();
     void trySwitchContextInProgressChanged();
     void loadedModelInfoChanged();
+    void generatedQuestionsChanged();
 
 private Q_SLOTS:
     void handleResponseChanged(const QString &response);
     void handleModelLoadingPercentageChanged(float);
     void promptProcessing();
+    void generatingQuestions();
     void responseStopped(qint64 promptResponseMs);
     void generatedNameChanged(const QString &name);
+    void generatedQuestionFinished(const QString &question);
     void handleRecalculating();
     void handleModelLoadingError(const QString &error);
     void handleTokenSpeedChanged(const QString &tokenSpeed);
@@ -179,6 +186,7 @@ private:
     QString m_fallbackReason;
     QString m_response;
     QList<QString> m_collections;
+    QList<QString> m_generatedQuestions;
     ChatModel *m_chatModel;
     bool m_responseInProgress = false;
     ResponseState m_responseState;

gpt4all-chat/chatllm.cpp

@@ -750,7 +750,7 @@ bool ChatLLM::promptInternal(const QList<QString> &collectionList, const QString
     if (!databaseResults.isEmpty()) {
         QStringList results;
         for (const ResultInfo &info : databaseResults)
-            results << u"Collection: %1\nPath: %2\nSnippet: %3"_s.arg(info.collection, info.path, info.text);
+            results << u"Collection: %1\nPath: %2\nExcerpt: %3"_s.arg(info.collection, info.path, info.text);
 
         // FIXME(jared): use a Jinja prompt template instead of hardcoded Alpaca-style localdocs template
         docsContext = u"### Context:\n%1\n\n"_s.arg(results.join("\n\n"));
@@ -797,7 +797,13 @@ bool ChatLLM::promptInternal(const QList<QString> &collectionList, const QString
         m_response = trimmed;
         emit responseChanged(QString::fromStdString(m_response));
     }
+
+    SuggestionMode mode = MySettings::globalInstance()->suggestionMode();
+    if (mode == SuggestionMode::On || (!databaseResults.isEmpty() && mode == SuggestionMode::LocalDocsOnly))
+        generateQuestions(elapsed);
+    else
         emit responseStopped(elapsed);
+
     m_pristineLoadedState = false;
     return true;
 }
@@ -875,13 +881,19 @@ void ChatLLM::generateName()
     if (!isModelLoaded())
         return;
 
+    const QString chatNamePrompt = MySettings::globalInstance()->modelChatNamePrompt(m_modelInfo);
+    if (chatNamePrompt.trimmed().isEmpty()) {
+        qWarning() << "ChatLLM: not generating chat name because prompt is empty";
+        return;
+    }
+
     auto promptTemplate = MySettings::globalInstance()->modelPromptTemplate(m_modelInfo);
     auto promptFunc = std::bind(&ChatLLM::handleNamePrompt, this, std::placeholders::_1);
     auto responseFunc = std::bind(&ChatLLM::handleNameResponse, this, std::placeholders::_1, std::placeholders::_2);
     auto recalcFunc = std::bind(&ChatLLM::handleNameRecalculate, this, std::placeholders::_1);
     LLModel::PromptContext ctx = m_ctx;
-    m_llModelInfo.model->prompt("Describe the above conversation in seven words or less.",
+    m_llModelInfo.model->prompt(chatNamePrompt.toStdString(), promptTemplate.toStdString(),
-                                promptTemplate.toStdString(), promptFunc, responseFunc, recalcFunc, ctx);
+                                promptFunc, responseFunc, recalcFunc, ctx);
     std::string trimmed = trim_whitespace(m_nameResponse);
     if (trimmed != m_nameResponse) {
         m_nameResponse = trimmed;
@@ -901,7 +913,6 @@ bool ChatLLM::handleNamePrompt(int32_t token)
     qDebug() << "name prompt" << m_llmThread.objectName() << token;
 #endif
     Q_UNUSED(token);
-    qt_noop();
     return !m_stopGenerating;
 }
 
@@ -925,10 +936,84 @@ bool ChatLLM::handleNameRecalculate(bool isRecalc)
     qDebug() << "name recalc" << m_llmThread.objectName() << isRecalc;
 #endif
     Q_UNUSED(isRecalc);
-    qt_noop();
     return true;
 }
 
+bool ChatLLM::handleQuestionPrompt(int32_t token)
+{
+#if defined(DEBUG)
+    qDebug() << "question prompt" << m_llmThread.objectName() << token;
+#endif
+    Q_UNUSED(token);
+    return !m_stopGenerating;
+}
+
+bool ChatLLM::handleQuestionResponse(int32_t token, const std::string &response)
+{
+#if defined(DEBUG)
+    qDebug() << "question response" << m_llmThread.objectName() << token << response;
+#endif
+    Q_UNUSED(token);
+
+    // add token to buffer
+    m_questionResponse.append(response);
+
+    // match whole question sentences
+    static const QRegularExpression reQuestion(R"(\b(What|Where|How|Why|When|Who|Which|Whose|Whom)\b[^?]*\?)");
+
+    // extract all questions from response
+    int lastMatchEnd = -1;
+    for (const auto &match : reQuestion.globalMatch(m_questionResponse)) {
+        lastMatchEnd = match.capturedEnd();
+        emit generatedQuestionFinished(match.captured());
+    }
+
+    // remove processed input from buffer
+    if (lastMatchEnd != -1)
+        m_questionResponse.erase(m_questionResponse.cbegin(), m_questionResponse.cbegin() + lastMatchEnd);
+
+    return true;
+}
+
+bool ChatLLM::handleQuestionRecalculate(bool isRecalc)
+{
+#if defined(DEBUG)
+    qDebug() << "name recalc" << m_llmThread.objectName() << isRecalc;
+#endif
+    Q_UNUSED(isRecalc);
+    return true;
+}
+
+void ChatLLM::generateQuestions(qint64 elapsed)
+{
+    Q_ASSERT(isModelLoaded());
+    if (!isModelLoaded()) {
+        emit responseStopped(elapsed);
+        return;
+    }
+
+    const std::string suggestedFollowUpPrompt = MySettings::globalInstance()->modelSuggestedFollowUpPrompt(m_modelInfo).toStdString();
+    if (QString::fromStdString(suggestedFollowUpPrompt).trimmed().isEmpty()) {
+        emit responseStopped(elapsed);
+        return;
+    }
+
+    emit generatingQuestions();
+    m_questionResponse.clear();
+    auto promptTemplate = MySettings::globalInstance()->modelPromptTemplate(m_modelInfo);
+    auto promptFunc = std::bind(&ChatLLM::handleQuestionPrompt, this, std::placeholders::_1);
+    auto responseFunc = std::bind(&ChatLLM::handleQuestionResponse, this, std::placeholders::_1, std::placeholders::_2);
+    auto recalcFunc = std::bind(&ChatLLM::handleQuestionRecalculate, this, std::placeholders::_1);
+    LLModel::PromptContext ctx = m_ctx;
+    QElapsedTimer totalTime;
+    totalTime.start();
+    m_llModelInfo.model->prompt(suggestedFollowUpPrompt,
+                                promptTemplate.toStdString(), promptFunc, responseFunc, recalcFunc, ctx);
+    elapsed += totalTime.elapsed();
+    emit responseStopped(elapsed);
+}
+
+
 bool ChatLLM::handleSystemPrompt(int32_t token)
 {
 #if defined(DEBUG)

gpt4all-chat/chatllm.h

@@ -160,6 +160,7 @@ public Q_SLOTS:
     void unloadModel();
     void reloadModel();
     void generateName();
+    void generateQuestions(qint64 elapsed);
     void handleChatIdChanged(const QString &id);
     void handleShouldBeLoadedChanged();
     void handleThreadStarted();
@@ -176,8 +177,10 @@ Q_SIGNALS:
     void modelLoadingWarning(const QString &warning);
     void responseChanged(const QString &response);
     void promptProcessing();
+    void generatingQuestions();
     void responseStopped(qint64 promptResponseMs);
     void generatedNameChanged(const QString &name);
+    void generatedQuestionFinished(const QString &generatedQuestion);
     void stateChanged();
     void threadStarted();
     void shouldBeLoadedChanged();
@@ -206,6 +209,9 @@ protected:
     bool handleRestoreStateFromTextPrompt(int32_t token);
     bool handleRestoreStateFromTextResponse(int32_t token, const std::string &response);
     bool handleRestoreStateFromTextRecalculate(bool isRecalc);
+    bool handleQuestionPrompt(int32_t token);
+    bool handleQuestionResponse(int32_t token, const std::string &response);
+    bool handleQuestionRecalculate(bool isRecalc);
     void saveState();
     void restoreState();
 
@@ -219,6 +225,7 @@ private:
 
     std::string m_response;
     std::string m_nameResponse;
+    QString m_questionResponse;
     LLModelInfo m_llModelInfo;
     LLModelType m_llModelType;
     ModelInfo m_modelInfo;

gpt4all-chat/icons/plus.svg

@@ -0,0 +1 @@
+<svg xmlns="http://www.w3.org/2000/svg" width="32" height="32" fill="#000000" viewBox="0 0 256 256"><path d="M224,128a8,8,0,0,1-8,8H136v80a8,8,0,0,1-16,0V136H40a8,8,0,0,1,0-16h80V40a8,8,0,0,1,16,0v80h80A8,8,0,0,1,224,128Z"></path></svg>

gpt4all-chat/icons/stack.svg

@@ -0,0 +1 @@
+<svg xmlns="http://www.w3.org/2000/svg" width="32" height="32" fill="#000000" viewBox="0 0 256 256"><path d="M230.91,124A8,8,0,0,1,228,134.91l-96,56a8,8,0,0,1-8.06,0l-96-56A8,8,0,0,1,36,121.09l92,53.65,92-53.65A8,8,0,0,1,230.91,124ZM24,80a8,8,0,0,1,4-6.91l96-56a8,8,0,0,1,8.06,0l96,56a8,8,0,0,1,0,13.82l-96,56a8,8,0,0,1-8.06,0l-96-56A8,8,0,0,1,24,80Zm23.88,0L128,126.74,208.12,80,128,33.26ZM232,192H216V176a8,8,0,0,0-16,0v16H184a8,8,0,0,0,0,16h16v16a8,8,0,0,0,16,0V208h16a8,8,0,0,0,0-16Zm-92,23.76-12,7L36,169.09A8,8,0,0,0,28,182.91l96,56a8,8,0,0,0,8.06,0l16-9.33A8,8,0,1,0,140,215.76Z"></path></svg>

gpt4all-chat/modellist.cpp

@@ -334,6 +334,28 @@ void ModelInfo::setSystemPrompt(const QString &p)
     m_systemPrompt = p;
 }
 
+QString ModelInfo::chatNamePrompt() const
+{
+    return MySettings::globalInstance()->modelChatNamePrompt(*this);
+}
+
+void ModelInfo::setChatNamePrompt(const QString &p)
+{
+    if (shouldSaveMetadata()) MySettings::globalInstance()->setModelChatNamePrompt(*this, p, true /*force*/);
+    m_chatNamePrompt = p;
+}
+
+QString ModelInfo::suggestedFollowUpPrompt() const
+{
+    return MySettings::globalInstance()->modelSuggestedFollowUpPrompt(*this);
+}
+
+void ModelInfo::setSuggestedFollowUpPrompt(const QString &p)
+{
+    if (shouldSaveMetadata()) MySettings::globalInstance()->setModelSuggestedFollowUpPrompt(*this, p, true /*force*/);
+    m_suggestedFollowUpPrompt = p;
+}
+
 bool ModelInfo::shouldSaveMetadata() const
 {
     return installed && (isClone() || isDiscovered() || description() == "" /*indicates sideloaded*/);
@@ -364,6 +386,8 @@ QVariantMap ModelInfo::getFields() const
         { "repeatPenaltyTokens", m_repeatPenaltyTokens },
         { "promptTemplate",      m_promptTemplate },
         { "systemPrompt",        m_systemPrompt },
+        { "chatNamePrompt",      m_chatNamePrompt },
+        { "suggestedFollowUpPrompt", m_suggestedFollowUpPrompt },
     };
 }
 
@@ -758,6 +782,10 @@ QVariant ModelList::dataInternal(const ModelInfo *info, int role) const
             return info->promptTemplate();
         case SystemPromptRole:
             return info->systemPrompt();
+        case ChatNamePromptRole:
+            return info->chatNamePrompt();
+        case SuggestedFollowUpPromptRole:
+            return info->suggestedFollowUpPrompt();
         case LikesRole:
             return info->likes();
         case DownloadsRole:
@@ -928,6 +956,10 @@ void ModelList::updateData(const QString &id, const QVector<QPair<int, QVariant>
                 info->setPromptTemplate(value.toString()); break;
             case SystemPromptRole:
                 info->setSystemPrompt(value.toString()); break;
+            case ChatNamePromptRole:
+                info->setChatNamePrompt(value.toString()); break;
+            case SuggestedFollowUpPromptRole:
+                info->setSuggestedFollowUpPrompt(value.toString()); break;
             case LikesRole:
                 {
                     if (info->likes() != value.toInt()) {
@@ -1077,6 +1109,8 @@ QString ModelList::clone(const ModelInfo &model)
         { ModelList::RepeatPenaltyTokensRole, model.repeatPenaltyTokens() },
         { ModelList::PromptTemplateRole, model.promptTemplate() },
         { ModelList::SystemPromptRole, model.systemPrompt() },
+        { ModelList::ChatNamePromptRole, model.chatNamePrompt() },
+        { ModelList::SuggestedFollowUpPromptRole, model.suggestedFollowUpPrompt() },
     };
     updateData(id, data);
     return id;
@@ -1772,6 +1806,14 @@ void ModelList::updateModelsFromSettings()
             const QString systemPrompt = settings.value(g + "/systemPrompt").toString();
             data.append({ ModelList::SystemPromptRole, systemPrompt });
         }
+        if (settings.contains(g + "/chatNamePrompt")) {
+            const QString chatNamePrompt = settings.value(g + "/chatNamePrompt").toString();
+            data.append({ ModelList::ChatNamePromptRole, chatNamePrompt });
+        }
+        if (settings.contains(g + "/suggestedFollowUpPrompt")) {
+            const QString suggestedFollowUpPrompt = settings.value(g + "/suggestedFollowUpPrompt").toString();
+            data.append({ ModelList::SuggestedFollowUpPromptRole, suggestedFollowUpPrompt });
+        }
         updateData(id, data);
     }
 }

gpt4all-chat/modellist.h

@@ -68,6 +68,8 @@ struct ModelInfo {
     Q_PROPERTY(int repeatPenaltyTokens READ repeatPenaltyTokens WRITE setRepeatPenaltyTokens)
     Q_PROPERTY(QString promptTemplate READ promptTemplate WRITE setPromptTemplate)
     Q_PROPERTY(QString systemPrompt READ systemPrompt WRITE setSystemPrompt)
+    Q_PROPERTY(QString chatNamePrompt READ chatNamePrompt WRITE setChatNamePrompt)
+    Q_PROPERTY(QString suggestedFollowUpPrompt READ suggestedFollowUpPrompt WRITE setSuggestedFollowUpPrompt)
     Q_PROPERTY(int likes READ likes WRITE setLikes)
     Q_PROPERTY(int downloads READ downloads WRITE setDownloads)
     Q_PROPERTY(QDateTime recency READ recency WRITE setRecency)
@@ -167,6 +169,10 @@ public:
     void setPromptTemplate(const QString &t);
     QString systemPrompt() const;
     void setSystemPrompt(const QString &p);
+    QString chatNamePrompt() const;
+    void setChatNamePrompt(const QString &p);
+    QString suggestedFollowUpPrompt() const;
+    void setSuggestedFollowUpPrompt(const QString &p);
 
     bool shouldSaveMetadata() const;
 
@@ -199,6 +205,8 @@ private:
     int     m_repeatPenaltyTokens     = 64;
     QString m_promptTemplate          = "### Human:\n%1\n\n### Assistant:\n";
     QString m_systemPrompt            = "### System:\nYou are an AI assistant who gives a quality response to whatever humans ask of you.\n\n";
+    QString m_chatNamePrompt          = "Describe the above conversation in seven words or less.";
+    QString m_suggestedFollowUpPrompt = "Suggest three very short factual follow-up questions that have not been answered yet or cannot be found inspired by the previous conversation and excerpts.";
     friend class MySettings;
 };
 Q_DECLARE_METATYPE(ModelInfo)
@@ -317,6 +325,8 @@ public:
         RepeatPenaltyTokensRole,
         PromptTemplateRole,
         SystemPromptRole,
+        ChatNamePromptRole,
+        SuggestedFollowUpPromptRole,
         MinPRole,
         LikesRole,
         DownloadsRole,
@@ -368,6 +378,8 @@ public:
         roles[RepeatPenaltyTokensRole] = "repeatPenaltyTokens";
         roles[PromptTemplateRole] = "promptTemplate";
         roles[SystemPromptRole] = "systemPrompt";
+        roles[ChatNamePromptRole] = "chatNamePrompt";
+        roles[SuggestedFollowUpPromptRole] = "suggestedFollowUpPrompt";
         roles[LikesRole] = "likes";
         roles[DownloadsRole] = "downloads";
         roles[RecencyRole] = "recency";

gpt4all-chat/mysettings.cpp

@@ -41,6 +41,7 @@ static const QVariantMap basicDefaults {
     { "saveChatsContext",         false },
     { "serverChat",               false },
     { "userDefaultModel",         "Application default" },
+    { "suggestionMode",           QVariant::fromValue(SuggestionMode::LocalDocsOnly) },
     { "localdocs/chunkSize",      512 },
     { "localdocs/retrievalSize",  3 },
     { "localdocs/showReferences", true },
@@ -136,6 +137,8 @@ void MySettings::restoreModelDefaults(const ModelInfo &info)
     setModelRepeatPenaltyTokens(info, info.m_repeatPenaltyTokens);
     setModelPromptTemplate(info, info.m_promptTemplate);
     setModelSystemPrompt(info, info.m_systemPrompt);
+    setModelChatNamePrompt(info, info.m_chatNamePrompt);
+    setModelSuggestedFollowUpPrompt(info, info.m_suggestedFollowUpPrompt);
 }
 
 void MySettings::restoreApplicationDefaults()
@@ -150,6 +153,7 @@ void MySettings::restoreApplicationDefaults()
     setModelPath(defaultLocalModelsPath());
     setUserDefaultModel(basicDefaults.value("userDefaultModel").toString());
     setForceMetal(defaults::forceMetal);
+    setSuggestionMode(basicDefaults.value("suggestionMode").value<SuggestionMode>());
 }
 
 void MySettings::restoreLocalDocsDefaults()
@@ -231,9 +235,11 @@ int       MySettings::modelPromptBatchSize    (const ModelInfo &info) const { re
 int       MySettings::modelContextLength          (const ModelInfo &info) const { return getModelSetting("contextLength",           info).toInt(); }
 int       MySettings::modelGpuLayers              (const ModelInfo &info) const { return getModelSetting("gpuLayers",               info).toInt(); }
 double    MySettings::modelRepeatPenalty          (const ModelInfo &info) const { return getModelSetting("repeatPenalty",           info).toDouble(); }
-int       MySettings::modelRepeatPenaltyTokens(const ModelInfo &info) const { return getModelSetting("repeatPenaltyTokens", info).toInt(); }
+int       MySettings::modelRepeatPenaltyTokens    (const ModelInfo &info) const { return getModelSetting("repeatPenaltyTokens",     info).toInt(); }
 QString   MySettings::modelPromptTemplate         (const ModelInfo &info) const { return getModelSetting("promptTemplate",          info).toString(); }
 QString   MySettings::modelSystemPrompt           (const ModelInfo &info) const { return getModelSetting("systemPrompt",            info).toString(); }
+QString   MySettings::modelChatNamePrompt         (const ModelInfo &info) const { return getModelSetting("chatNamePrompt",          info).toString(); }
+QString   MySettings::modelSuggestedFollowUpPrompt(const ModelInfo &info) const { return getModelSetting("suggestedFollowUpPrompt", info).toString(); }
 
 void MySettings::setModelFilename(const ModelInfo &info, const QString &value, bool force)
 {
@@ -345,6 +351,16 @@ void MySettings::setModelSystemPrompt(const ModelInfo &info, const QString &valu
     setModelSetting("systemPrompt", info, value, force, true);
 }
 
+void MySettings::setModelChatNamePrompt(const ModelInfo &info, const QString &value, bool force)
+{
+    setModelSetting("chatNamePrompt", info, value, force, true);
+}
+
+void MySettings::setModelSuggestedFollowUpPrompt(const ModelInfo &info, const QString &value, bool force)
+{
+    setModelSetting("suggestedFollowUpPrompt", info, value, force, true);
+}
+
 int MySettings::threadCount() const
 {
     int c = m_settings.value("threadCount", defaults::threadCount).toInt();
@@ -383,6 +399,7 @@ bool        MySettings::localDocsUseRemoteEmbed() const { return getBasicSetting
 QString        MySettings::localDocsNomicAPIKey() const    { return getBasicSetting("localdocs/nomicAPIKey"   ).toString(); }
 QString        MySettings::localDocsEmbedDevice() const    { return getBasicSetting("localdocs/embedDevice"   ).toString(); }
 QString        MySettings::networkAttribution() const      { return getBasicSetting("network/attribution"     ).toString(); }
+SuggestionMode MySettings::suggestionMode() const          { return getBasicSetting("suggestionMode").value<SuggestionMode>(); };
 
 void MySettings::setSaveChatsContext(bool value)                      { setBasicSetting("saveChatsContext",         value); }
 void MySettings::setServerChat(bool value)                            { setBasicSetting("serverChat",               value); }
@@ -399,6 +416,7 @@ void MySettings::setLocalDocsUseRemoteEmbed(bool value)               { setBasic
 void MySettings::setLocalDocsNomicAPIKey(const QString &value)        { setBasicSetting("localdocs/nomicAPIKey",    value, "localDocsNomicAPIKey"); }
 void MySettings::setLocalDocsEmbedDevice(const QString &value)        { setBasicSetting("localdocs/embedDevice",    value, "localDocsEmbedDevice"); }
 void MySettings::setNetworkAttribution(const QString &value)          { setBasicSetting("network/attribution",      value, "networkAttribution"); }
+void MySettings::setSuggestionMode(SuggestionMode value)              { setBasicSetting("suggestionMode",           int(value)); }
 
 QString MySettings::modelPath()
 {

gpt4all-chat/mysettings.h

@@ -13,6 +13,18 @@
 #include <cstdint>
 #include <optional>
 
+namespace MySettingsEnums {
+    Q_NAMESPACE
+
+    enum class SuggestionMode {
+        LocalDocsOnly = 0,
+        On = 1,
+        Off = 2,
+    };
+    Q_ENUM_NS(SuggestionMode)
+}
+using namespace MySettingsEnums;
+
 class MySettings : public QObject
 {
     Q_OBJECT
@@ -39,6 +51,7 @@ class MySettings : public QObject
     Q_PROPERTY(QStringList deviceList MEMBER m_deviceList CONSTANT)
     Q_PROPERTY(QStringList embeddingsDeviceList MEMBER m_embeddingsDeviceList CONSTANT)
     Q_PROPERTY(int networkPort READ networkPort WRITE setNetworkPort NOTIFY networkPortChanged)
+    Q_PROPERTY(SuggestionMode suggestionMode READ suggestionMode WRITE setSuggestionMode NOTIFY suggestionModeChanged)
 
 public:
     static MySettings *globalInstance();
@@ -98,6 +111,10 @@ public:
     Q_INVOKABLE void setModelContextLength(const ModelInfo &info, int value, bool force = false);
     int modelGpuLayers(const ModelInfo &info) const;
     Q_INVOKABLE void setModelGpuLayers(const ModelInfo &info, int value, bool force = false);
+    QString modelChatNamePrompt(const ModelInfo &info) const;
+    Q_INVOKABLE void setModelChatNamePrompt(const ModelInfo &info, const QString &value, bool force = false);
+    QString modelSuggestedFollowUpPrompt(const ModelInfo &info) const;
+    Q_INVOKABLE void setModelSuggestedFollowUpPrompt(const ModelInfo &info, const QString &value, bool force = false);
 
     // Application settings
     int threadCount() const;
@@ -122,6 +139,8 @@ public:
     void setContextLength(int32_t value);
     int32_t gpuLayers() const;
     void setGpuLayers(int32_t value);
+    SuggestionMode suggestionMode() const;
+    void setSuggestionMode(SuggestionMode mode);
 
     // Release/Download settings
     QString lastVersionStarted() const;
@@ -171,6 +190,8 @@ Q_SIGNALS:
     void repeatPenaltyTokensChanged(const ModelInfo &info);
     void promptTemplateChanged(const ModelInfo &info);
     void systemPromptChanged(const ModelInfo &info);
+    void chatNamePromptChanged(const ModelInfo &info);
+    void suggestedFollowUpPromptChanged(const ModelInfo &info);
     void threadCountChanged();
     void saveChatsContextChanged();
     void serverChatChanged();
@@ -193,6 +214,7 @@ Q_SIGNALS:
     void networkUsageStatsActiveChanged();
     void attemptModelLoadChanged();
     void deviceChanged();
+    void suggestionModeChanged();
 
 private:
     QSettings m_settings;

gpt4all-chat/qml/ApplicationSettings.qml

@@ -227,16 +227,40 @@ MySettingsTab {
                 MySettings.userDefaultModel = comboBox.currentText
             }
         }
+        MySettingsLabel {
+            id: suggestionModeLabel
+            text: qsTr("Suggestion Mode")
+            helpText: qsTr("Generate suggested follow-up questions at the end of responses.")
+            Layout.row: 6
+            Layout.column: 0
+        }
+        MyComboBox {
+            id: suggestionModeBox
+            Layout.row: 6
+            Layout.column: 2
+            Layout.minimumWidth: 400
+            Layout.maximumWidth: 400
+            Layout.alignment: Qt.AlignRight
+            model: [ qsTr("When chatting with LocalDocs"), qsTr("Whenever possible"), qsTr("Never") ]
+            Accessible.name: suggestionModeLabel.text
+            Accessible.description: suggestionModeLabel.helpText
+            onActivated: {
+                MySettings.suggestionMode = suggestionModeBox.currentIndex;
+            }
+            Component.onCompleted: {
+                suggestionModeBox.currentIndex = MySettings.suggestionMode;
+            }
+        }
         MySettingsLabel {
             id: modelPathLabel
             text: qsTr("Download Path")
             helpText: qsTr("Where to store local models and the LocalDocs database.")
-            Layout.row: 6
+            Layout.row: 7
             Layout.column: 0
         }
 
         RowLayout {
-            Layout.row: 6
+            Layout.row: 7
             Layout.column: 2
             Layout.alignment: Qt.AlignRight
             Layout.minimumWidth: 400
@@ -273,12 +297,12 @@ MySettingsTab {
             id: dataLakeLabel
             text: qsTr("Enable Datalake")
             helpText: qsTr("Send chats and feedback to the GPT4All Open-Source Datalake.")
-            Layout.row: 7
+            Layout.row: 8
             Layout.column: 0
         }
         MyCheckBox {
             id: dataLakeBox
-            Layout.row: 7
+            Layout.row: 8
             Layout.column: 2
             Layout.alignment: Qt.AlignRight
             Component.onCompleted: { dataLakeBox.checked = MySettings.networkIsActive; }
@@ -296,7 +320,7 @@ MySettingsTab {
         }
 
         ColumnLayout {
-            Layout.row: 8
+            Layout.row: 9
             Layout.column: 0
             Layout.columnSpan: 3
             Layout.fillWidth: true
@@ -319,7 +343,7 @@ MySettingsTab {
             id: nThreadsLabel
             text: qsTr("CPU Threads")
             helpText: qsTr("The number of CPU threads used for inference and embedding.")
-            Layout.row: 9
+            Layout.row: 10
             Layout.column: 0
         }
         MyTextField {
@@ -327,7 +351,7 @@ MySettingsTab {
             color: theme.textColor
             font.pixelSize: theme.fontSizeLarge
             Layout.alignment: Qt.AlignRight
-            Layout.row: 9
+            Layout.row: 10
             Layout.column: 2
             Layout.minimumWidth: 200
             Layout.maximumWidth: 200
@@ -351,12 +375,12 @@ MySettingsTab {
             id: saveChatsContextLabel
             text: qsTr("Save Chat Context")
             helpText: qsTr("Save the chat model's state to disk for faster loading. WARNING: Uses ~2GB per chat.")
-            Layout.row: 10
+            Layout.row: 11
             Layout.column: 0
         }
         MyCheckBox {
             id: saveChatsContextBox
-            Layout.row: 10
+            Layout.row: 11
             Layout.column: 2
             Layout.alignment: Qt.AlignRight
             checked: MySettings.saveChatsContext
@@ -368,12 +392,12 @@ MySettingsTab {
             id: serverChatLabel
             text: qsTr("Enable Local Server")
             helpText: qsTr("Expose an OpenAI-Compatible server to localhost. WARNING: Results in increased resource usage.")
-            Layout.row: 11
+            Layout.row: 12
             Layout.column: 0
         }
         MyCheckBox {
             id: serverChatBox
-            Layout.row: 11
+            Layout.row: 12
             Layout.column: 2
             Layout.alignment: Qt.AlignRight
             checked: MySettings.serverChat
@@ -385,7 +409,7 @@ MySettingsTab {
             id: serverPortLabel
             text: qsTr("API Server Port")
             helpText: qsTr("The port to use for the local server. Requires restart.")
-            Layout.row: 12
+            Layout.row: 13
             Layout.column: 0
         }
         MyTextField {
@@ -393,7 +417,7 @@ MySettingsTab {
             text: MySettings.networkPort
             color: theme.textColor
             font.pixelSize: theme.fontSizeLarge
-            Layout.row: 12
+            Layout.row: 13
             Layout.column: 2
             Layout.minimumWidth: 200
             Layout.maximumWidth: 200

gpt4all-chat/qml/ChatView.qml

@@ -797,7 +797,7 @@ Rectangle {
 
                             delegate: GridLayout {
                                 width: listView.contentItem.width - 15
-                                rows: 3
+                                rows: 5
                                 columns: 2
 
                                 Item {
@@ -850,6 +850,8 @@ Rectangle {
                                             font.pixelSize: theme.fontSizeLarger
                                             font.bold: true
                                             color: theme.conversationHeader
+                                            enabled: false
+                                            focus: false
                                             readOnly: true
                                         }
                                         Text {
@@ -872,6 +874,7 @@ Rectangle {
                                                     case Chat.LocalDocsProcessing: return qsTr("searching localdocs: ") + currentChat.collectionList.join(", ") + " ...";
                                                     case Chat.PromptProcessing: return qsTr("processing ...")
                                                     case Chat.ResponseGeneration: return qsTr("generating response ...");
+                                                    case Chat.GeneratingQuestions: return qsTr("generating questions ...");
                                                     default: return ""; // handle unexpected values
                                                     }
                                                 }
@@ -1094,7 +1097,16 @@ Rectangle {
                                     Layout.alignment: Qt.AlignVCenter
                                     Layout.preferredWidth: childrenRect.width
                                     Layout.preferredHeight: childrenRect.height
-                                    visible: consolidatedSources.length !== 0 && MySettings.localDocsShowReferences && (!currentResponse || !currentChat.responseInProgress)
+                                    visible: {
+                                        if (consolidatedSources.length === 0)
+                                            return false
+                                        if (!MySettings.localDocsShowReferences)
+                                            return false
+                                        if (currentResponse && currentChat.responseInProgress
+                                            && currentChat.responseState !== Chat.GeneratingQuestions )
+                                            return false
+                                        return true
+                                    }
 
                                     MyButton {
                                         backgroundColor: theme.sourcesBackground
@@ -1171,7 +1183,16 @@ Rectangle {
                                     Layout.row: 3
                                     Layout.column: 1
                                     Layout.topMargin: 5
-                                    visible: consolidatedSources.length !== 0 && MySettings.localDocsShowReferences && (!currentResponse || !currentChat.responseInProgress)
+                                    visible: {
+                                        if (consolidatedSources.length === 0)
+                                            return false
+                                        if (!MySettings.localDocsShowReferences)
+                                            return false
+                                        if (currentResponse && currentChat.responseInProgress
+                                            && currentChat.responseState !== Chat.GeneratingQuestions )
+                                            return false
+                                        return true
+                                    }
                                     clip: true
                                     Layout.fillWidth: true
                                     Layout.preferredHeight: 0
@@ -1310,45 +1331,250 @@ Rectangle {
                                         }
                                     }
                                 }
+
+                                function shouldShowSuggestions() {
+                                    if (!currentResponse)
+                                        return false;
+                                    if (MySettings.suggestionMode === 2) // Off
+                                        return false;
+                                    if (MySettings.suggestionMode === 0 && consolidatedSources.length === 0) // LocalDocs only
+                                        return false;
+                                    return currentChat.responseState === Chat.GeneratingQuestions || currentChat.generatedQuestions.length !== 0;
                                 }
 
-                            property bool shouldAutoScroll: true
+                                Item {
-                            property bool isAutoScrolling: false
+                                    visible: shouldShowSuggestions()
+                                    Layout.row: 4
+                                    Layout.column: 0
+                                    Layout.topMargin: 20
+                                    Layout.alignment: Qt.AlignVCenter | Qt.AlignRight
+                                    Layout.preferredWidth: 28
+                                    Layout.preferredHeight: 28
+                                    Image {
+                                        id: stack
+                                        sourceSize: Qt.size(28, 28)
+                                        fillMode: Image.PreserveAspectFit
+                                        mipmap: true
+                                        visible: false
+                                        source: "qrc:/gpt4all/icons/stack.svg"
+                                    }
 
-                            Connections {
+                                    ColorOverlay {
-                                target: currentChat
+                                        anchors.fill: stack
-                                function onResponseChanged() {
+                                        source: stack
-                                    listView.scrollToEnd()
+                                        color: theme.conversationHeader
+                                    }
+                                }
+
+                                Item {
+                                    visible: shouldShowSuggestions()
+                                    Layout.row: 4
+                                    Layout.column: 1
+                                    Layout.topMargin: 20
+                                    Layout.fillWidth: true
+                                    Layout.preferredHeight: 38
+                                    RowLayout {
+                                        spacing: 5
+                                        anchors.left: parent.left
+                                        anchors.top: parent.top
+                                        anchors.bottom: parent.bottom
+
+                                        TextArea {
+                                            text: qsTr("Suggested follow-ups")
+                                            padding: 0
+                                            font.pixelSize: theme.fontSizeLarger
+                                            font.bold: true
+                                            color: theme.conversationHeader
+                                            enabled: false
+                                            focus: false
+                                            readOnly: true
+                                        }
+                                    }
+                                }
+
+                                ColumnLayout {
+                                    visible: shouldShowSuggestions()
+                                    Layout.row: 5
+                                    Layout.column: 1
+                                    Layout.fillWidth: true
+                                    Layout.minimumHeight: 1
+                                    spacing: 10
+                                    Repeater {
+                                        model: currentChat.generatedQuestions
+                                        TextArea {
+                                            id: followUpText
+                                            Layout.fillWidth: true
+                                            Layout.alignment: Qt.AlignLeft
+                                            rightPadding: 40
+                                            topPadding: 10
+                                            leftPadding: 20
+                                            bottomPadding: 10
+                                            text: modelData
+                                            focus: false
+                                            readOnly: true
+                                            wrapMode: Text.WordWrap
+                                            hoverEnabled: !currentChat.responseInProgress
+                                            color: theme.textColor
+                                            font.pixelSize: theme.fontSizeLarge
+                                            background: Rectangle {
+                                                color: hovered ? theme.sourcesBackgroundHovered : theme.sourcesBackground
+                                                radius: 10
+                                            }
+                                            MouseArea {
+                                                id: maFollowUp
+                                                anchors.fill: parent
+                                                enabled: !currentChat.responseInProgress
+                                                onClicked: function() {
+                                                    var chat = window.currentChat
+                                                    var followup = modelData
+                                                    chat.stopGenerating()
+                                                    chat.newPromptResponsePair(followup);
+                                                    chat.prompt(followup,
+                                                                MySettings.promptTemplate,
+                                                                MySettings.maxLength,
+                                                                MySettings.topK,
+                                                                MySettings.topP,
+                                                                MySettings.minP,
+                                                                MySettings.temperature,
+                                                                MySettings.promptBatchSize,
+                                                                MySettings.repeatPenalty,
+                                                                MySettings.repeatPenaltyTokens)
+                                                }
+                                            }
+                                            Item {
+                                                anchors.right: parent.right
+                                                anchors.verticalCenter: parent.verticalCenter
+                                                width: 40
+                                                height: 40
+                                                visible: !currentChat.responseInProgress
+                                                Image {
+                                                    id: plusImage
+                                                    anchors.verticalCenter: parent.verticalCenter
+                                                    sourceSize.width: 20
+                                                    sourceSize.height: 20
+                                                    mipmap: true
+                                                    visible: false
+                                                    source: "qrc:/gpt4all/icons/plus.svg"
+                                                }
+
+                                                ColorOverlay {
+                                                    anchors.fill: plusImage
+                                                    source: plusImage
+                                                    color: theme.styledTextColor
+                                                }
+                                            }
+                                        }
+                                    }
+
+                                    Rectangle {
+                                        Layout.fillWidth: true
+                                        color: "transparent"
+                                        radius: 10
+                                        Layout.preferredHeight: currentChat.responseInProgress ? 40 : 0
+                                        clip: true
+                                        ColumnLayout {
+                                            id: followUpLayout
+                                            anchors.fill: parent
+                                            Rectangle {
+                                                id: myRect1
+                                                Layout.preferredWidth: 0
+                                                Layout.minimumWidth: 0
+                                                Layout.maximumWidth: parent.width
+                                                height: 12
+                                                color: theme.sourcesBackgroundHovered
+                                            }
+
+                                            Rectangle {
+                                                id: myRect2
+                                                Layout.preferredWidth: 0
+                                                Layout.minimumWidth: 0
+                                                Layout.maximumWidth: parent.width
+                                                height: 12
+                                                color: theme.sourcesBackgroundHovered
+                                            }
+
+                                            SequentialAnimation {
+                                                id: followUpProgressAnimation
+                                                ParallelAnimation {
+                                                    PropertyAnimation {
+                                                        target: myRect1
+                                                        property: "Layout.preferredWidth"
+                                                        from: 0
+                                                        to: followUpLayout.width
+                                                        duration: 1000
+                                                    }
+                                                    PropertyAnimation {
+                                                        target: myRect2
+                                                        property: "Layout.preferredWidth"
+                                                        from: 0
+                                                        to: followUpLayout.width / 2
+                                                        duration: 1000
+                                                    }
+                                                }
+                                                SequentialAnimation {
+                                                    loops: Animation.Infinite
+                                                    ParallelAnimation {
+                                                        PropertyAnimation {
+                                                            target: myRect1
+                                                            property: "opacity"
+                                                            from: 1
+                                                            to: 0.2
+                                                            duration: 1500
+                                                        }
+                                                        PropertyAnimation {
+                                                            target: myRect2
+                                                            property: "opacity"
+                                                            from: 1
+                                                            to: 0.2
+                                                            duration: 1500
+                                                        }
+                                                    }
+                                                    ParallelAnimation {
+                                                        PropertyAnimation {
+                                                            target: myRect1
+                                                            property: "opacity"
+                                                            from: 0.2
+                                                            to: 1
+                                                            duration: 1500
+                                                        }
+                                                        PropertyAnimation {
+                                                            target: myRect2
+                                                            property: "opacity"
+                                                            from: 0.2
+                                                            to: 1
+                                                            duration: 1500
+                                                        }
+                                                    }
+                                                }
+                                            }
+
+                                            onVisibleChanged: {
+                                                if (visible)
+                                                    followUpProgressAnimation.start();
+                                            }
+                                        }
+
+                                        Behavior on Layout.preferredHeight {
+                                            NumberAnimation {
+                                                duration: 300
+                                                easing.type: Easing.InOutQuad
+                                            }
+                                        }
+                                    }
                                 }
                             }
 
                             function scrollToEnd() {
-                                if (listView.shouldAutoScroll) {
-                                    listView.isAutoScrolling = true
                                 listView.positionViewAtEnd()
-                                    listView.isAutoScrolling = false
-                                }
                             }
 
-                            onContentYChanged: {
+                            onContentHeightChanged: {
-                                if (!isAutoScrolling)
+                                if (atYEnd)
-                                    shouldAutoScroll = atYEnd
+                                    scrollToEnd()
-                            }
-
-                            Component.onCompleted: {
-                                shouldAutoScroll = true
-                                positionViewAtEnd()
-                            }
-
-                            footer: Item {
-                                id: bottomPadding
-                                width: parent.width
-                                height: 0
                             }
                         }
                     }
                 }
-
             }
 
             Rectangle {

gpt4all-chat/qml/ModelSettings.qml

@@ -250,45 +250,64 @@ MySettingsTab {
             }
         }
 
+        MySettingsLabel {
+            id: chatNamePromptLabel
+            text: qsTr("Chat Name Prompt")
+            helpText: qsTr("Prompt used to automatically generate chat names.")
+            Layout.row: 11
+            Layout.column: 0
+            Layout.topMargin: 15
+        }
+
         Rectangle {
-            id: optionalImageRect
+            id: chatNamePrompt
-            visible: false // FIXME: for later
+            Layout.row: 12
-            Layout.row: 2
+            Layout.column: 0
-            Layout.column: 1
+            Layout.columnSpan: 2
-            Layout.rowSpan: 5
+            Layout.fillWidth: true
-            Layout.alignment: Qt.AlignHCenter
+            Layout.minimumHeight: Math.max(100, chatNamePromptTextArea.contentHeight + 20)
-            Layout.fillHeight: true
-            Layout.maximumWidth: height
-            Layout.topMargin: 35
-            Layout.bottomMargin: 35
-            Layout.leftMargin: 35
-            width: 3000
-            radius: 10
             color: "transparent"
-            Item {
+            clip: true
-                anchors.centerIn: parent
+            MyTextArea {
-                height: childrenRect.height
+                id: chatNamePromptTextArea
-                Image {
+                anchors.fill: parent
-                    id: img
+                text: root.currentModelInfo.chatNamePrompt
-                    anchors.horizontalCenter: parent.horizontalCenter
+                Accessible.role: Accessible.EditableText
-                    width: 100
+                Accessible.name: chatNamePromptLabel.text
-                    height: 100
+                Accessible.description: chatNamePromptLabel.text
-                    source: "qrc:/gpt4all/icons/image.svg"
             }
-                Text {
-                    text: qsTr("Add\noptional image")
-                    font.pixelSize: theme.fontSizeLarge
-                    anchors.top: img.bottom
-                    anchors.horizontalCenter: parent.horizontalCenter
-                    wrapMode: TextArea.Wrap
-                    horizontalAlignment: Qt.AlignHCenter
-                    color: theme.mutedTextColor
         }
+
+        MySettingsLabel {
+            id: suggestedFollowUpPromptLabel
+            text: qsTr("Suggested FollowUp Prompt")
+            helpText: qsTr("Prompt used to generate suggested follow-up questions.")
+            Layout.row: 13
+            Layout.column: 0
+            Layout.topMargin: 15
+        }
+
+        Rectangle {
+            id: suggestedFollowUpPrompt
+            Layout.row: 14
+            Layout.column: 0
+            Layout.columnSpan: 2
+            Layout.fillWidth: true
+            Layout.minimumHeight: Math.max(100, suggestedFollowUpPromptTextArea.contentHeight + 20)
+            color: "transparent"
+            clip: true
+            MyTextArea {
+                id: suggestedFollowUpPromptTextArea
+                anchors.fill: parent
+                text: root.currentModelInfo.suggestedFollowUpPrompt
+                Accessible.role: Accessible.EditableText
+                Accessible.name: suggestedFollowUpPromptLabel.text
+                Accessible.description: suggestedFollowUpPromptLabel.text
             }
         }
 
         GridLayout {
-            Layout.row: 11
+            Layout.row: 15
             Layout.column: 0
             Layout.columnSpan: 2
             Layout.topMargin: 15
@@ -784,7 +803,7 @@ MySettingsTab {
         }
 
         Rectangle {
-            Layout.row: 12
+            Layout.row: 16
             Layout.column: 0
             Layout.columnSpan: 2
             Layout.topMargin: 15