From 6d9cdf228c890e8e54064b60cb7eb4f3c4efe081 Mon Sep 17 00:00:00 2001 From: Adam Treat Date: Sat, 1 Jul 2023 11:34:21 -0400 Subject: [PATCH] Huge change that completely revamps the settings dialog and implements per model settings as well as the ability to clone a model into a "character." This also implements system prompts as well as quite a few bugfixes for instance this fixes chatgpt. --- gpt4all-chat/CMakeLists.txt | 2 +- gpt4all-chat/chat.cpp | 50 +- gpt4all-chat/chat.h | 9 +- gpt4all-chat/chatlistmodel.cpp | 2 +- gpt4all-chat/chatllm.cpp | 111 +++- gpt4all-chat/chatllm.h | 12 +- gpt4all-chat/download.cpp | 2 +- gpt4all-chat/icons/image.svg | 6 + gpt4all-chat/main.qml | 36 +- gpt4all-chat/metadata/models.json | 4 +- gpt4all-chat/modellist.cpp | 597 ++++++++++++++++--- gpt4all-chat/modellist.h | 112 +++- gpt4all-chat/mysettings.cpp | 232 ++++++-- gpt4all-chat/mysettings.h | 71 +-- gpt4all-chat/network.cpp | 6 +- gpt4all-chat/qml/ApplicationSettings.qml | 49 +- gpt4all-chat/qml/GenerationSettings.qml | 302 ---------- gpt4all-chat/qml/LocalDocsSettings.qml | 345 +++++------ gpt4all-chat/qml/ModelDownloaderDialog.qml | 9 +- gpt4all-chat/qml/ModelSettings.qml | 662 ++++++++++++++++++++- gpt4all-chat/qml/MyButton.qml | 2 +- gpt4all-chat/qml/MyComboBox.qml | 3 +- gpt4all-chat/qml/MyDirectoryField.qml | 2 +- gpt4all-chat/qml/MySettingsStack.qml | 101 ++-- gpt4all-chat/qml/MySettingsTab.qml | 73 ++- gpt4all-chat/qml/MyTextField.qml | 2 +- gpt4all-chat/qml/SettingsDialog.qml | 11 +- gpt4all-chat/qml/Theme.qml | 6 +- gpt4all-chat/responsetext.cpp | 68 +++ gpt4all-chat/server.cpp | 35 +- 30 files changed, 2041 insertions(+), 881 deletions(-) create mode 100644 gpt4all-chat/icons/image.svg delete mode 100644 gpt4all-chat/qml/GenerationSettings.qml diff --git a/gpt4all-chat/CMakeLists.txt b/gpt4all-chat/CMakeLists.txt index f47375a6..84480739 100644 --- a/gpt4all-chat/CMakeLists.txt +++ b/gpt4all-chat/CMakeLists.txt @@ -99,7 +99,6 @@ qt_add_qml_module(chat qml/AboutDialog.qml qml/Theme.qml qml/ModelSettings.qml - qml/GenerationSettings.qml qml/ApplicationSettings.qml qml/LocalDocsSettings.qml qml/MySettingsTab.qml @@ -119,6 +118,7 @@ qt_add_qml_module(chat icons/db.svg icons/settings.svg icons/edit.svg + icons/image.svg icons/trash.svg icons/network.svg icons/thumbs_up.svg diff --git a/gpt4all-chat/chat.cpp b/gpt4all-chat/chat.cpp index 5b8e198e..ea63879c 100644 --- a/gpt4all-chat/chat.cpp +++ b/gpt4all-chat/chat.cpp @@ -67,6 +67,7 @@ void Chat::connectLLM() connect(this, &Chat::regenerateResponseRequested, m_llmodel, &ChatLLM::regenerateResponse, Qt::QueuedConnection); connect(this, &Chat::resetResponseRequested, m_llmodel, &ChatLLM::resetResponse, Qt::QueuedConnection); connect(this, &Chat::resetContextRequested, m_llmodel, &ChatLLM::resetContext, Qt::QueuedConnection); + connect(this, &Chat::processSystemPromptRequested, m_llmodel, &ChatLLM::processSystemPrompt, Qt::QueuedConnection); connect(ModelList::globalInstance()->installedModels(), &InstalledModels::countChanged, this, &Chat::handleModelInstalled, Qt::QueuedConnection); @@ -93,6 +94,11 @@ void Chat::reset() m_chatModel->clear(); } +void Chat::processSystemPrompt() +{ + emit processSystemPromptRequested(); +} + bool Chat::isModelLoaded() const { return m_isModelLoaded; @@ -111,27 +117,10 @@ void Chat::resetResponseState() emit responseStateChanged(); } -void Chat::prompt(const QString &prompt, const QString &prompt_template, int32_t n_predict, - int32_t top_k, float top_p, float temp, int32_t n_batch, float repeat_penalty, - int32_t repeat_penalty_tokens) +void Chat::prompt(const QString &prompt) { resetResponseState(); - int threadCount = MySettings::globalInstance()->threadCount(); - if (threadCount <= 0) - threadCount = std::min(4, (int32_t) std::thread::hardware_concurrency()); - - emit promptRequested( - m_collections, - prompt, - prompt_template, - n_predict, - top_k, - top_p, - temp, - n_batch, - repeat_penalty, - repeat_penalty_tokens, - threadCount); + emit promptRequested( m_collections, prompt); } void Chat::regenerateResponse() @@ -386,7 +375,10 @@ bool Chat::serialize(QDataStream &stream, int version) const stream << m_id; stream << m_name; stream << m_userName; - stream << m_modelInfo.filename; + if (version > 4) + stream << m_modelInfo.id(); + else + stream << m_modelInfo.filename(); if (version > 2) stream << m_collections; if (!m_llmodel->serialize(stream, version)) @@ -405,16 +397,22 @@ bool Chat::deserialize(QDataStream &stream, int version) stream >> m_userName; emit nameChanged(); - QString filename; - stream >> filename; - if (!ModelList::globalInstance()->contains(filename)) - return false; - m_modelInfo = ModelList::globalInstance()->modelInfo(filename); + QString modelId; + stream >> modelId; + if (version > 4) { + if (!ModelList::globalInstance()->contains(modelId)) + return false; + m_modelInfo = ModelList::globalInstance()->modelInfo(modelId); + } else { + if (!ModelList::globalInstance()->containsByFilename(modelId)) + return false; + m_modelInfo = ModelList::globalInstance()->modelInfoByFilename(modelId); + } emit modelInfoChanged(); // Prior to version 2 gptj models had a bug that fixed the kv_cache to F32 instead of F16 so // unfortunately, we cannot deserialize these - if (version < 2 && m_modelInfo.filename.contains("gpt4all-j")) + if (version < 2 && m_modelInfo.filename().contains("gpt4all-j")) return false; if (version > 2) { stream >> m_collections; diff --git a/gpt4all-chat/chat.h b/gpt4all-chat/chat.h index 834abb28..22988684 100644 --- a/gpt4all-chat/chat.h +++ b/gpt4all-chat/chat.h @@ -53,9 +53,9 @@ public: ChatModel *chatModel() { return m_chatModel; } Q_INVOKABLE void reset(); + Q_INVOKABLE void processSystemPrompt(); Q_INVOKABLE bool isModelLoaded() const; - Q_INVOKABLE void prompt(const QString &prompt, const QString &prompt_template, int32_t n_predict, - int32_t top_k, float top_p, float temp, int32_t n_batch, float repeat_penalty, int32_t repeat_penalty_tokens); + Q_INVOKABLE void prompt(const QString &prompt); Q_INVOKABLE void regenerateResponse(); Q_INVOKABLE void stopGenerating(); Q_INVOKABLE void newPromptResponsePair(const QString &prompt); @@ -102,12 +102,11 @@ Q_SIGNALS: void responseChanged(); void responseInProgressChanged(); void responseStateChanged(); - void promptRequested(const QList &collectionList, const QString &prompt, const QString &prompt_template, - int32_t n_predict, int32_t top_k, float top_p, float temp, int32_t n_batch, float repeat_penalty, - int32_t repeat_penalty_tokens, int32_t n_threads); + void promptRequested(const QList &collectionList, const QString &prompt); void regenerateResponseRequested(); void resetResponseRequested(); void resetContextRequested(); + void processSystemPromptRequested(); void modelChangeRequested(const ModelInfo &modelInfo); void modelInfoChanged(); void recalcChanged(); diff --git a/gpt4all-chat/chatlistmodel.cpp b/gpt4all-chat/chatlistmodel.cpp index fc401326..062ebdd9 100644 --- a/gpt4all-chat/chatlistmodel.cpp +++ b/gpt4all-chat/chatlistmodel.cpp @@ -5,7 +5,7 @@ #include #define CHAT_FORMAT_MAGIC 0xF5D553CC -#define CHAT_FORMAT_VERSION 4 +#define CHAT_FORMAT_VERSION 5 class MyChatListModel: public ChatListModel { }; Q_GLOBAL_STATIC(MyChatListModel, chatListModelInstance) diff --git a/gpt4all-chat/chatllm.cpp b/gpt4all-chat/chatllm.cpp index 14f3ce6f..065946b1 100644 --- a/gpt4all-chat/chatllm.cpp +++ b/gpt4all-chat/chatllm.cpp @@ -69,6 +69,7 @@ ChatLLM::ChatLLM(Chat *parent, bool isServer) , m_isServer(isServer) , m_forceMetal(MySettings::globalInstance()->forceMetal()) , m_reloadingToChangeVariant(false) + , m_processedSystemPrompt(false) { moveToThread(&m_llmThread); connect(this, &ChatLLM::sendStartup, Network::globalInstance(), &Network::sendStartup); @@ -123,7 +124,7 @@ void ChatLLM::handleForceMetalChanged(bool forceMetal) bool ChatLLM::loadDefaultModel() { ModelInfo defaultModel = ModelList::globalInstance()->defaultModelInfo(); - if (defaultModel.filename.isEmpty()) { + if (defaultModel.filename().isEmpty()) { emit modelLoadingError(QString("Could not find any model to load")); return false; } @@ -145,7 +146,7 @@ bool ChatLLM::loadModel(const ModelInfo &modelInfo) return true; bool isChatGPT = modelInfo.isChatGPT; - QString filePath = modelInfo.dirpath + modelInfo.filename; + QString filePath = modelInfo.dirpath + modelInfo.filename(); QFileInfo fileInfo(filePath); // We have a live model, but it isn't the one we want @@ -186,6 +187,11 @@ bool ChatLLM::loadModel(const ModelInfo &modelInfo) #endif restoreState(); emit isModelLoadedChanged(true); + Q_ASSERT(!m_modelInfo.filename().isEmpty()); + if (m_modelInfo.filename().isEmpty()) + emit modelLoadingError(QString("Modelinfo is left null for %1").arg(modelInfo.filename())); + else + processSystemPrompt(); return true; } else { // Release the memory since we have to switch to a different model. @@ -237,7 +243,7 @@ bool ChatLLM::loadModel(const ModelInfo &modelInfo) if (!m_isServer) LLModelStore::globalInstance()->releaseModel(m_llModelInfo); // release back into the store m_llModelInfo = LLModelInfo(); - emit modelLoadingError(QString("Could not load model due to invalid model file for %1").arg(modelInfo.filename)); + emit modelLoadingError(QString("Could not load model due to invalid model file for %1").arg(modelInfo.filename())); } else { switch (m_llModelInfo.model->implementation().modelType[0]) { case 'L': m_llModelType = LLModelType::LLAMA_; break; @@ -251,7 +257,7 @@ bool ChatLLM::loadModel(const ModelInfo &modelInfo) if (!m_isServer) LLModelStore::globalInstance()->releaseModel(m_llModelInfo); // release back into the store m_llModelInfo = LLModelInfo(); - emit modelLoadingError(QString("Could not determine model type for %1").arg(modelInfo.filename)); + emit modelLoadingError(QString("Could not determine model type for %1").arg(modelInfo.filename())); } } } @@ -259,7 +265,7 @@ bool ChatLLM::loadModel(const ModelInfo &modelInfo) if (!m_isServer) LLModelStore::globalInstance()->releaseModel(m_llModelInfo); // release back into the store m_llModelInfo = LLModelInfo(); - emit modelLoadingError(QString("Could not load model due to invalid format for %1").arg(modelInfo.filename)); + emit modelLoadingError(QString("Could not load model due to invalid format for %1").arg(modelInfo.filename())); } } #if defined(DEBUG_MODEL_LOADING) @@ -282,12 +288,13 @@ bool ChatLLM::loadModel(const ModelInfo &modelInfo) if (!m_isServer) LLModelStore::globalInstance()->releaseModel(m_llModelInfo); // release back into the store m_llModelInfo = LLModelInfo(); - emit modelLoadingError(QString("Could not find file for model %1").arg(modelInfo.filename)); + emit modelLoadingError(QString("Could not find file for model %1").arg(modelInfo.filename())); } if (m_llModelInfo.model) setModelInfo(modelInfo); + processSystemPrompt(); return m_llModelInfo.model; } @@ -323,6 +330,7 @@ void ChatLLM::resetResponse() void ChatLLM::resetContext() { regenerateResponse(); + m_processedSystemPrompt = false; m_ctx = LLModel::PromptContext(); } @@ -412,9 +420,25 @@ bool ChatLLM::handleRecalculate(bool isRecalc) } return !m_stopGenerating; } +bool ChatLLM::prompt(const QList &collectionList, const QString &prompt) +{ + if (!m_processedSystemPrompt) + processSystemPrompt(); + const QString promptTemplate = MySettings::globalInstance()->modelPromptTemplate(m_modelInfo); + const int32_t n_predict = MySettings::globalInstance()->modelMaxLength(m_modelInfo); + const int32_t top_k = MySettings::globalInstance()->modelTopK(m_modelInfo); + const float top_p = MySettings::globalInstance()->modelTopP(m_modelInfo); + const float temp = MySettings::globalInstance()->modelTemperature(m_modelInfo); + const int32_t n_batch = MySettings::globalInstance()->modelPromptBatchSize(m_modelInfo); + const float repeat_penalty = MySettings::globalInstance()->modelRepeatPenalty(m_modelInfo); + const int32_t repeat_penalty_tokens = MySettings::globalInstance()->modelRepeatPenaltyTokens(m_modelInfo); + return promptInternal(collectionList, prompt, promptTemplate, n_predict, top_k, top_p, temp, n_batch, + repeat_penalty, repeat_penalty_tokens); +} -bool ChatLLM::prompt(const QList &collectionList, const QString &prompt, const QString &prompt_template, int32_t n_predict, int32_t top_k, - float top_p, float temp, int32_t n_batch, float repeat_penalty, int32_t repeat_penalty_tokens, int n_threads) +bool ChatLLM::promptInternal(const QList &collectionList, const QString &prompt, const QString &promptTemplate, + int32_t n_predict, int32_t top_k, float top_p, float temp, int32_t n_batch, float repeat_penalty, + int32_t repeat_penalty_tokens) { if (!isModelLoaded()) return false; @@ -430,10 +454,14 @@ bool ChatLLM::prompt(const QList &collectionList, const QString &prompt augmentedTemplate.append("### Context:"); for (const ResultInfo &info : databaseResults) augmentedTemplate.append(info.text); - augmentedTemplate.append(prompt_template); + augmentedTemplate.append(promptTemplate); QString instructPrompt = augmentedTemplate.join("\n").arg(prompt); + int n_threads = MySettings::globalInstance()->threadCount(); + if (n_threads <= 0) + n_threads = std::min(4, (int32_t) std::thread::hardware_concurrency()); + m_stopGenerating = false; auto promptFunc = std::bind(&ChatLLM::handlePrompt, this, std::placeholders::_1); auto responseFunc = std::bind(&ChatLLM::handleResponse, this, std::placeholders::_1, @@ -515,7 +543,7 @@ void ChatLLM::reloadModel() qDebug() << "reloadModel" << m_llmThread.objectName() << m_llModelInfo.model; #endif const ModelInfo m = modelInfo(); - if (m.name.isEmpty()) + if (m.name().isEmpty()) loadDefaultModel(); else loadModel(m); @@ -581,6 +609,27 @@ bool ChatLLM::handleNameRecalculate(bool isRecalc) return true; } +bool ChatLLM::handleSystemPrompt(int32_t token) +{ + Q_UNUSED(token); + qt_noop(); + return true; +} + +bool ChatLLM::handleSystemResponse(int32_t token, const std::string &response) +{ + Q_UNUSED(token); + Q_UNUSED(response); + return false; +} + +bool ChatLLM::handleSystemRecalculate(bool isRecalc) +{ + Q_UNUSED(isRecalc); + Q_UNREACHABLE(); + return true; +} + bool ChatLLM::serialize(QDataStream &stream, int version) { if (version > 1) { @@ -701,3 +750,45 @@ void ChatLLM::restoreState() m_state.clear(); m_state.resize(0); } + +void ChatLLM::processSystemPrompt() +{ + Q_ASSERT(isModelLoaded()); + if (!isModelLoaded() || m_processedSystemPrompt || m_isServer) + return; + + auto promptFunc = std::bind(&ChatLLM::handleSystemPrompt, this, std::placeholders::_1); + auto responseFunc = std::bind(&ChatLLM::handleSystemResponse, this, std::placeholders::_1, + std::placeholders::_2); + auto recalcFunc = std::bind(&ChatLLM::handleSystemRecalculate, this, std::placeholders::_1); + + const std::string systemPrompt = MySettings::globalInstance()->modelSystemPrompt(m_modelInfo).toStdString(); + const int32_t n_predict = MySettings::globalInstance()->modelMaxLength(m_modelInfo); + const int32_t top_k = MySettings::globalInstance()->modelTopK(m_modelInfo); + const float top_p = MySettings::globalInstance()->modelTopP(m_modelInfo); + const float temp = MySettings::globalInstance()->modelTemperature(m_modelInfo); + const int32_t n_batch = MySettings::globalInstance()->modelPromptBatchSize(m_modelInfo); + const float repeat_penalty = MySettings::globalInstance()->modelRepeatPenalty(m_modelInfo); + const int32_t repeat_penalty_tokens = MySettings::globalInstance()->modelRepeatPenaltyTokens(m_modelInfo); + int n_threads = MySettings::globalInstance()->threadCount(); + if (n_threads <= 0) + n_threads = std::min(4, (int32_t) std::thread::hardware_concurrency()); + m_ctx.n_predict = n_predict; + m_ctx.top_k = top_k; + m_ctx.top_p = top_p; + m_ctx.temp = temp; + m_ctx.n_batch = n_batch; + m_ctx.repeat_penalty = repeat_penalty; + m_ctx.repeat_last_n = repeat_penalty_tokens; + m_llModelInfo.model->setThreadCount(n_threads); +#if defined(DEBUG) + printf("%s", qPrintable(QString::fromStdString(systemPrompt))); + fflush(stdout); +#endif + m_llModelInfo.model->prompt(systemPrompt, promptFunc, responseFunc, recalcFunc, m_ctx); +#if defined(DEBUG) + printf("\n"); + fflush(stdout); +#endif + m_processedSystemPrompt = true; +} \ No newline at end of file diff --git a/gpt4all-chat/chatllm.h b/gpt4all-chat/chatllm.h index 9a1c17b5..aad8a0fb 100644 --- a/gpt4all-chat/chatllm.h +++ b/gpt4all-chat/chatllm.h @@ -97,9 +97,7 @@ public: bool deserialize(QDataStream &stream, int version); public Q_SLOTS: - bool prompt(const QList &collectionList, const QString &prompt, const QString &prompt_template, - int32_t n_predict, int32_t top_k, float top_p, float temp, int32_t n_batch, float repeat_penalty, - int32_t repeat_penalty_tokens, int32_t n_threads); + bool prompt(const QList &collectionList, const QString &prompt); bool loadDefaultModel(); bool loadModel(const ModelInfo &modelInfo); void modelChangeRequested(const ModelInfo &modelInfo); @@ -111,6 +109,7 @@ public Q_SLOTS: void handleShouldBeLoadedChanged(); void handleThreadStarted(); void handleForceMetalChanged(bool forceMetal); + void processSystemPrompt(); Q_SIGNALS: void recalcChanged(); @@ -131,12 +130,18 @@ Q_SIGNALS: void modelInfoChanged(const ModelInfo &modelInfo); protected: + bool promptInternal(const QList &collectionList, const QString &prompt, const QString &promptTemplate, + int32_t n_predict, int32_t top_k, float top_p, float temp, int32_t n_batch, float repeat_penalty, + int32_t repeat_penalty_tokens); bool handlePrompt(int32_t token); bool handleResponse(int32_t token, const std::string &response); bool handleRecalculate(bool isRecalc); bool handleNamePrompt(int32_t token); bool handleNameResponse(int32_t token, const std::string &response); bool handleNameRecalculate(bool isRecalc); + bool handleSystemPrompt(int32_t token); + bool handleSystemResponse(int32_t token, const std::string &response); + bool handleSystemRecalculate(bool isRecalc); void saveState(); void restoreState(); @@ -160,6 +165,7 @@ private: bool m_isServer; bool m_forceMetal; bool m_reloadingToChangeVariant; + bool m_processedSystemPrompt; }; #endif // CHATLLM_H diff --git a/gpt4all-chat/download.cpp b/gpt4all-chat/download.cpp index 858747e2..dc8f81dc 100644 --- a/gpt4all-chat/download.cpp +++ b/gpt4all-chat/download.cpp @@ -175,7 +175,7 @@ void Download::installModel(const QString &modelFile, const QString &apiKey) return; Network::globalInstance()->sendInstallModel(modelFile); - QString filePath = MySettings::globalInstance()->modelPath() + modelFile + ".txt"; + QString filePath = MySettings::globalInstance()->modelPath() + modelFile; QFile file(filePath); if (file.open(QIODeviceBase::WriteOnly | QIODeviceBase::Text)) { QTextStream stream(&file); diff --git a/gpt4all-chat/icons/image.svg b/gpt4all-chat/icons/image.svg new file mode 100644 index 00000000..f4c46f14 --- /dev/null +++ b/gpt4all-chat/icons/image.svg @@ -0,0 +1,6 @@ + + + diff --git a/gpt4all-chat/main.qml b/gpt4all-chat/main.qml index 12aa9e45..83ad8300 100644 --- a/gpt4all-chat/main.qml +++ b/gpt4all-chat/main.qml @@ -204,12 +204,12 @@ Window { anchors.horizontalCenterOffset: window.width >= 950 ? 0 : Math.max(-((950 - window.width) / 2), -99.5) enabled: !currentChat.isServer model: ModelList.installedModels - valueRole: "filename" + valueRole: "id" textRole: "name" property string currentModelName: "" function updateCurrentModelName() { - var info = ModelList.modelInfo(currentChat.modelInfo.filename); - comboBox.currentModelName = info.name !== "" ? info.name : info.filename; + var info = ModelList.modelInfo(currentChat.modelInfo.id); + comboBox.currentModelName = info.name; } Connections { target: currentChat @@ -217,6 +217,16 @@ Window { comboBox.updateCurrentModelName(); } } + Connections { + target: window + function onCurrentChatChanged() { + comboBox.updateCurrentModelName(); + } + } + background: Rectangle { + color: theme.backgroundDark + radius: 10 + } contentItem: Text { anchors.horizontalCenter: parent.horizontalCenter leftPadding: 10 @@ -233,7 +243,7 @@ Window { delegate: ItemDelegate { width: comboBox.width contentItem: Text { - text: name !== "" ? name : filename + text: name color: theme.textColor font: comboBox.font elide: Text.ElideRight @@ -557,6 +567,7 @@ Window { onClicked: { Network.sendResetContext(chatModel.count) currentChat.reset(); + currentChat.processSystemPrompt(); } } @@ -870,6 +881,7 @@ Window { } MyButton { + id: myButton visible: chatModel.count && !currentChat.isServer Image { anchors.verticalCenter: parent.verticalCenter @@ -894,19 +906,17 @@ Window { chatModel.updateThumbsUpState(index, false); chatModel.updateThumbsDownState(index, false); chatModel.updateNewResponse(index, ""); - currentChat.prompt(listElement.prompt, - MySettings.promptTemplate, - MySettings.maxLength, - MySettings.topK, - MySettings.topP, - MySettings.temperature, - MySettings.promptBatchSize, - MySettings.repeatPenalty, - MySettings.repeatPenaltyTokens) + currentChat.prompt(listElement.prompt) } } } } + background: Rectangle { + border.color: myButton.down ? theme.backgroundLightest : theme.buttonBorder + border.width: 2 + radius: 10 + color: myButton.hovered ? theme.backgroundLighter : theme.backgroundLight + } anchors.bottom: textInputView.top anchors.horizontalCenter: textInputView.horizontalCenter anchors.bottomMargin: 20 diff --git a/gpt4all-chat/metadata/models.json b/gpt4all-chat/metadata/models.json index fe38a91d..0a259de3 100644 --- a/gpt4all-chat/metadata/models.json +++ b/gpt4all-chat/metadata/models.json @@ -64,7 +64,9 @@ "parameters": "7 billion", "quant": "q4_0", "type": "MPT", - "description": "Best overall smaller model
  • Fast responses
  • Chat based
  • Trained by Mosaic ML
  • Cannot be used commercially
" + "description": "Best overall smaller model
  • Fast responses
  • Chat based
  • Trained by Mosaic ML
  • Cannot be used commercially
", + "promptTemplate": "<|im_start|>user\n%1<|im_end|><|im_start|>assistant\n", + "systemPrompt": "<|im_start|>system\n- You are a helpful assistant chatbot trained by MosaicML.\n- You answer questions.\n- You are excited to be able to help the user, but will refuse to do anything that could be considered harmful to the user.\n- You are more than just an information source, you are also able to write poetry, short stories, and make jokes.<|im_end|>" }, { "order": "g", diff --git a/gpt4all-chat/modellist.cpp b/gpt4all-chat/modellist.cpp index 4193784b..c1f5a089 100644 --- a/gpt4all-chat/modellist.cpp +++ b/gpt4all-chat/modellist.cpp @@ -1,10 +1,142 @@ #include "modellist.h" #include "mysettings.h" +#include "network.h" #include //#define USE_LOCAL_MODELSJSON +QString ModelInfo::id() const +{ + return m_id; +} + +void ModelInfo::setId(const QString &id) +{ + m_id = id; +} + +QString ModelInfo::name() const +{ + return MySettings::globalInstance()->modelName(*this); +} + +void ModelInfo::setName(const QString &name) +{ + if (isClone) MySettings::globalInstance()->setModelName(*this, name, isClone /*force*/); + m_name = name; +} + +QString ModelInfo::filename() const +{ + return MySettings::globalInstance()->modelFilename(*this); +} + +void ModelInfo::setFilename(const QString &filename) +{ + if (isClone) MySettings::globalInstance()->setModelFilename(*this, filename, isClone /*force*/); + m_filename = filename; +} + +double ModelInfo::temperature() const +{ + return MySettings::globalInstance()->modelTemperature(*this); +} + +void ModelInfo::setTemperature(double t) +{ + if (isClone) MySettings::globalInstance()->setModelTemperature(*this, t, isClone /*force*/); + m_temperature = t; +} + +double ModelInfo::topP() const +{ + return MySettings::globalInstance()->modelTopP(*this); +} + +void ModelInfo::setTopP(double p) +{ + if (isClone) MySettings::globalInstance()->setModelTopP(*this, p, isClone /*force*/); + m_topP = p; +} + +int ModelInfo::topK() const +{ + return MySettings::globalInstance()->modelTopK(*this); +} + +void ModelInfo::setTopK(int k) +{ + if (isClone) MySettings::globalInstance()->setModelTopK(*this, k, isClone /*force*/); + m_topK = k; +} + +int ModelInfo::maxLength() const +{ + return MySettings::globalInstance()->modelMaxLength(*this); +} + +void ModelInfo::setMaxLength(int l) +{ + if (isClone) MySettings::globalInstance()->setModelMaxLength(*this, l, isClone /*force*/); + m_maxLength = l; +} + +int ModelInfo::promptBatchSize() const +{ + return MySettings::globalInstance()->modelPromptBatchSize(*this); +} + +void ModelInfo::setPromptBatchSize(int s) +{ + if (isClone) MySettings::globalInstance()->setModelPromptBatchSize(*this, s, isClone /*force*/); + m_promptBatchSize = s; +} + +double ModelInfo::repeatPenalty() const +{ + return MySettings::globalInstance()->modelRepeatPenalty(*this); +} + +void ModelInfo::setRepeatPenalty(double p) +{ + if (isClone) MySettings::globalInstance()->setModelRepeatPenalty(*this, p, isClone /*force*/); + m_repeatPenalty = p; +} + +int ModelInfo::repeatPenaltyTokens() const +{ + return MySettings::globalInstance()->modelRepeatPenaltyTokens(*this); +} + +void ModelInfo::setRepeatPenaltyTokens(int t) +{ + if (isClone) MySettings::globalInstance()->setModelRepeatPenaltyTokens(*this, t, isClone /*force*/); + m_repeatPenaltyTokens = t; +} + +QString ModelInfo::promptTemplate() const +{ + return MySettings::globalInstance()->modelPromptTemplate(*this); +} + +void ModelInfo::setPromptTemplate(const QString &t) +{ + if (isClone) MySettings::globalInstance()->setModelPromptTemplate(*this, t, isClone /*force*/); + m_promptTemplate = t; +} + +QString ModelInfo::systemPrompt() const +{ + return MySettings::globalInstance()->modelSystemPrompt(*this); +} + +void ModelInfo::setSystemPrompt(const QString &p) +{ + if (isClone) MySettings::globalInstance()->setModelSystemPrompt(*this, p, isClone /*force*/); + m_systemPrompt = p; +} + InstalledModels::InstalledModels(QObject *parent) : QSortFilterProxyModel(parent) { @@ -27,11 +159,11 @@ int InstalledModels::count() const return rowCount(); } -QString InstalledModels::firstFilename() const +QString InstalledModels::firstId() const { if (rowCount() > 0) { QModelIndex firstIndex = index(0, 0); - return sourceModel()->data(firstIndex, ModelList::FilenameRole).toString(); + return sourceModel()->data(firstIndex, ModelList::IdRole).toString(); } else { return QString(); } @@ -96,9 +228,23 @@ ModelList::ModelList() m_watcher->addPath(exePath); m_watcher->addPath(MySettings::globalInstance()->modelPath()); connect(m_watcher, &QFileSystemWatcher::directoryChanged, this, &ModelList::updateModelsFromDirectory); - connect(MySettings::globalInstance(), &MySettings::modelPathChanged, this, &ModelList::updateModelList); + connect(MySettings::globalInstance(), &MySettings::modelPathChanged, this, &ModelList::updateModelsFromDirectory); + connect(MySettings::globalInstance(), &MySettings::modelPathChanged, this, &ModelList::updateModelsFromJson); + connect(MySettings::globalInstance(), &MySettings::modelPathChanged, this, &ModelList::updateModelsFromSettings); + connect(MySettings::globalInstance(), &MySettings::nameChanged, this, &ModelList::updateDataForSettings); + connect(MySettings::globalInstance(), &MySettings::temperatureChanged, this, &ModelList::updateDataForSettings); + connect(MySettings::globalInstance(), &MySettings::topPChanged, this, &ModelList::updateDataForSettings); + connect(MySettings::globalInstance(), &MySettings::topKChanged, this, &ModelList::updateDataForSettings); + connect(MySettings::globalInstance(), &MySettings::maxLengthChanged, this, &ModelList::updateDataForSettings); + connect(MySettings::globalInstance(), &MySettings::promptBatchSizeChanged, this, &ModelList::updateDataForSettings); + connect(MySettings::globalInstance(), &MySettings::repeatPenaltyChanged, this, &ModelList::updateDataForSettings); + connect(MySettings::globalInstance(), &MySettings::repeatPenaltyTokensChanged, this, &ModelList::updateDataForSettings);; + connect(MySettings::globalInstance(), &MySettings::promptTemplateChanged, this, &ModelList::updateDataForSettings); + connect(MySettings::globalInstance(), &MySettings::systemPromptChanged, this, &ModelList::updateDataForSettings); + + updateModelsFromJson(); + updateModelsFromSettings(); updateModelsFromDirectory(); - updateModelList(); } QString ModelList::incompleteDownloadPath(const QString &modelFile) @@ -124,19 +270,19 @@ const QList ModelList::userDefaultModelList() const QList models; bool foundUserDefault = false; for (ModelInfo *info : m_models) { - if (info->installed && (info->name == userDefaultModelName || info->filename == userDefaultModelName)) { + if (info->installed && info->id() == userDefaultModelName) { foundUserDefault = true; - models.prepend(info->name.isEmpty() ? info->filename : info->name); + models.prepend(info->name()); } else if (info->installed) { - models.append(info->name.isEmpty() ? info->filename : info->name); + models.append(info->name()); } } - const QString defaultFileName = "Application default"; + const QString defaultId = "Application default"; if (foundUserDefault) - models.append(defaultFileName); + models.append(defaultId); else - models.prepend(defaultFileName); + models.prepend(defaultId); return models; } @@ -166,11 +312,11 @@ ModelInfo ModelList::defaultModelInfo() const break; // If we don't have a user specified default, but *do* have a default setting and match, then use it - if (!hasUserDefaultName && hasDefaultName && (defaultModel->name == defaultModelName || defaultModel->filename == defaultModelName)) + if (!hasUserDefaultName && hasDefaultName && (defaultModel->id() == defaultModelName)) break; // If we have a user specified default and match, then use it - if (hasUserDefaultName && (defaultModel->name == userDefaultModelName || defaultModel->filename == userDefaultModelName)) + if (hasUserDefaultName && (defaultModel->id() == userDefaultModelName)) break; } if (defaultModel) @@ -178,14 +324,28 @@ ModelInfo ModelList::defaultModelInfo() const return ModelInfo(); } -bool ModelList::contains(const QString &filename) const +bool ModelList::contains(const QString &id) const { QMutexLocker locker(&m_mutex); - return m_modelMap.contains(filename); + return m_modelMap.contains(id); +} + +bool ModelList::containsByFilename(const QString &filename) const +{ + QMutexLocker locker(&m_mutex); + for (ModelInfo *info : m_models) + if (info->filename() == filename) + return true; + return false; } bool ModelList::lessThan(const ModelInfo* a, const ModelInfo* b) { + // Rule 0: Non-clone before clone + if (a->isClone != b->isClone) { + return !a->isClone; + } + // Rule 1: Non-empty 'order' before empty if (a->order.isEmpty() != b->order.isEmpty()) { return !a->order.isEmpty(); @@ -196,27 +356,32 @@ bool ModelList::lessThan(const ModelInfo* a, const ModelInfo* b) return a->order < b->order; } - // Rule 3: Both 'order' are empty, sort by filename - return a->filename < b->filename; + // Rule 3: Both 'order' are empty, sort by id + return a->id() < b->id(); } -void ModelList::addModel(const QString &filename) +void ModelList::addModel(const QString &id) { - const bool hasModel = contains(filename); + const bool hasModel = contains(id); Q_ASSERT(!hasModel); if (hasModel) { - qWarning() << "ERROR: model list already contains" << filename; + qWarning() << "ERROR: model list already contains" << id; return; } - beginInsertRows(QModelIndex(), m_models.size(), m_models.size()); + int modelSizeBefore = 0; int modelSizeAfter = 0; + { + QMutexLocker locker(&m_mutex); + modelSizeBefore = m_models.size(); + } + beginInsertRows(QModelIndex(), modelSizeBefore, modelSizeBefore); { QMutexLocker locker(&m_mutex); ModelInfo *info = new ModelInfo; - info->filename = filename; + info->setId(id); m_models.append(info); - m_modelMap.insert(filename, info); + m_modelMap.insert(id, info); std::stable_sort(m_models.begin(), m_models.end(), ModelList::lessThan); modelSizeAfter = m_models.size(); } @@ -235,10 +400,12 @@ int ModelList::rowCount(const QModelIndex &parent) const QVariant ModelList::dataInternal(const ModelInfo *info, int role) const { switch (role) { + case IdRole: + return info->id(); case NameRole: - return info->name; + return info->name(); case FilenameRole: - return info->filename; + return info->filename(); case DirpathRole: return info->dirpath; case FilesizeRole: @@ -287,15 +454,35 @@ QVariant ModelList::dataInternal(const ModelInfo *info, int role) const return info->quant; case TypeRole: return info->type; + case IsCloneRole: + return info->isClone; + case TemperatureRole: + return info->temperature(); + case TopPRole: + return info->topP(); + case TopKRole: + return info->topK(); + case MaxLengthRole: + return info->maxLength(); + case PromptBatchSizeRole: + return info->promptBatchSize(); + case RepeatPenaltyRole: + return info->repeatPenalty(); + case RepeatPenaltyTokensRole: + return info->repeatPenaltyTokens(); + case PromptTemplateRole: + return info->promptTemplate(); + case SystemPromptRole: + return info->systemPrompt(); } return QVariant(); } -QVariant ModelList::data(const QString &filename, int role) const +QVariant ModelList::data(const QString &id, int role) const { QMutexLocker locker(&m_mutex); - ModelInfo *info = m_modelMap.value(filename); + ModelInfo *info = m_modelMap.value(id); return dataInternal(info, role); } @@ -308,7 +495,7 @@ QVariant ModelList::data(const QModelIndex &index, int role) const return dataInternal(info, role); } -void ModelList::updateData(const QString &filename, int role, const QVariant &value) +void ModelList::updateData(const QString &id, int role, const QVariant &value) { int modelSize; bool updateInstalled; @@ -316,23 +503,25 @@ void ModelList::updateData(const QString &filename, int role, const QVariant &va int index; { QMutexLocker locker(&m_mutex); - if (!m_modelMap.contains(filename)) { - qWarning() << "ERROR: cannot update as model map does not contain" << filename; + if (!m_modelMap.contains(id)) { + qWarning() << "ERROR: cannot update as model map does not contain" << id; return; } - ModelInfo *info = m_modelMap.value(filename); + ModelInfo *info = m_modelMap.value(id); index = m_models.indexOf(info); if (index == -1) { - qWarning() << "ERROR: cannot update as model list does not contain" << filename; + qWarning() << "ERROR: cannot update as model list does not contain" << id; return; } switch (role) { + case IdRole: + info->setId(value.toString()); break; case NameRole: - info->name = value.toString(); break; + info->setName(value.toString()); break; case FilenameRole: - info->filename = value.toString(); break; + info->setFilename(value.toString()); break; case DirpathRole: info->dirpath = value.toString(); break; case FilesizeRole: @@ -381,15 +570,35 @@ void ModelList::updateData(const QString &filename, int role, const QVariant &va info->quant = value.toString(); break; case TypeRole: info->type = value.toString(); break; + case IsCloneRole: + info->isClone = value.toBool(); break; + case TemperatureRole: + info->setTemperature(value.toDouble()); break; + case TopPRole: + info->setTopP(value.toDouble()); break; + case TopKRole: + info->setTopK(value.toInt()); break; + case MaxLengthRole: + info->setMaxLength(value.toInt()); break; + case PromptBatchSizeRole: + info->setPromptBatchSize(value.toInt()); break; + case RepeatPenaltyRole: + info->setRepeatPenalty(value.toDouble()); break; + case RepeatPenaltyTokensRole: + info->setRepeatPenaltyTokens(value.toInt()); break; + case PromptTemplateRole: + info->setPromptTemplate(value.toString()); break; + case SystemPromptRole: + info->setSystemPrompt(value.toString()); break; } // Extra guarantee that these always remains in sync with filesystem - QFileInfo fileInfo(info->dirpath + info->filename); + QFileInfo fileInfo(info->dirpath + info->filename()); if (info->installed != fileInfo.exists()) { info->installed = fileInfo.exists(); updateInstalled = true; } - QFileInfo incompleteInfo(incompleteDownloadPath(info->filename)); + QFileInfo incompleteInfo(incompleteDownloadPath(info->filename())); if (info->isIncomplete != incompleteInfo.exists()) { info->isIncomplete = incompleteInfo.exists(); updateIncomplete = true; @@ -402,12 +611,120 @@ void ModelList::updateData(const QString &filename, int role, const QVariant &va emit userDefaultModelListChanged(); } -ModelInfo ModelList::modelInfo(const QString &filename) const +ModelInfo ModelList::modelInfo(const QString &id) const { QMutexLocker locker(&m_mutex); - if (!m_modelMap.contains(filename)) + if (!m_modelMap.contains(id)) return ModelInfo(); - return *m_modelMap.value(filename); + return *m_modelMap.value(id); +} + +ModelInfo ModelList::modelInfoByFilename(const QString &filename) const +{ + QMutexLocker locker(&m_mutex); + for (ModelInfo *info : m_models) + if (info->filename() == filename) + return *info; + return ModelInfo(); +} + +bool ModelList::isUniqueName(const QString &name) const +{ + QMutexLocker locker(&m_mutex); + for (const ModelInfo *info : m_models) { + if(info->name() == name) + return false; + } + return true; +} + +QString ModelList::clone(const ModelInfo &model) +{ + const QString id = Network::globalInstance()->generateUniqueId(); + addModel(id); + updateData(id, ModelList::IsCloneRole, true); + updateData(id, ModelList::NameRole, uniqueModelName(model)); + updateData(id, ModelList::FilenameRole, model.filename()); + updateData(id, ModelList::DirpathRole, model.dirpath); + updateData(id, ModelList::InstalledRole, model.installed); + updateData(id, ModelList::ChatGPTRole, model.isChatGPT); + updateData(id, ModelList::TemperatureRole, model.temperature()); + updateData(id, ModelList::TopPRole, model.topP()); + updateData(id, ModelList::TopKRole, model.topK()); + updateData(id, ModelList::MaxLengthRole, model.maxLength()); + updateData(id, ModelList::PromptBatchSizeRole, model.promptBatchSize()); + updateData(id, ModelList::RepeatPenaltyRole, model.repeatPenalty()); + updateData(id, ModelList::RepeatPenaltyTokensRole, model.repeatPenaltyTokens()); + updateData(id, ModelList::PromptTemplateRole, model.promptTemplate()); + updateData(id, ModelList::SystemPromptRole, model.systemPrompt()); + return id; +} + +void ModelList::remove(const ModelInfo &model) +{ + Q_ASSERT(model.isClone); + if (!model.isClone) + return; + + const bool hasModel = contains(model.id()); + Q_ASSERT(hasModel); + if (!hasModel) { + qWarning() << "ERROR: model list does not contain" << model.id(); + return; + } + + int indexOfModel = 0; + int modelSizeAfter = 0; + { + QMutexLocker locker(&m_mutex); + ModelInfo *info = m_modelMap.value(model.id()); + indexOfModel = m_models.indexOf(info); + } + beginRemoveRows(QModelIndex(), indexOfModel, indexOfModel); + { + QMutexLocker locker(&m_mutex); + ModelInfo *info = m_models.takeAt(indexOfModel); + m_modelMap.remove(info->id()); + delete info; + modelSizeAfter = m_models.size(); + } + endRemoveRows(); + emit dataChanged(index(0, 0), index(modelSizeAfter - 1, 0)); + emit userDefaultModelListChanged(); + MySettings::globalInstance()->eraseModel(model); +} + +QString ModelList::uniqueModelName(const ModelInfo &model) const +{ + QMutexLocker locker(&m_mutex); + QRegularExpression re("^(.*)~(\\d+)$"); + QRegularExpressionMatch match = re.match(model.name()); + QString baseName; + if (match.hasMatch()) + baseName = match.captured(1); + else + baseName = model.name(); + + int maxSuffixNumber = 0; + bool baseNameExists = false; + + for (const ModelInfo *info : m_models) { + if(info->name() == baseName) + baseNameExists = true; + + QRegularExpressionMatch match = re.match(info->name()); + if (match.hasMatch()) { + QString currentBaseName = match.captured(1); + int currentSuffixNumber = match.captured(2).toInt(); + if (currentBaseName == baseName && currentSuffixNumber > maxSuffixNumber) + maxSuffixNumber = currentSuffixNumber; + } + } + + if (baseNameExists) + return baseName + "~" + QString::number(maxSuffixNumber + 1); + + return baseName; } QString ModelList::modelDirPath(const QString &modelName, bool isChatGPT) @@ -456,12 +773,25 @@ void ModelList::updateModelsFromDirectory() if (!info.exists()) continue; - if (!contains(filename)) - addModel(filename); + QVector modelsById; + { + QMutexLocker locker(&m_mutex); + for (ModelInfo *info : m_models) + if (info->filename() == filename) + modelsById.append(info->id()); + } - updateData(filename, ChatGPTRole, filename.startsWith("chatgpt-")); - updateData(filename, DirpathRole, path); - updateData(filename, FilesizeRole, toFileSize(info.size())); + if (modelsById.isEmpty()) { + addModel(filename); + modelsById.append(filename); + } + + for (const QString &id : modelsById) { + updateData(id, FilenameRole, filename); + updateData(id, ChatGPTRole, filename.startsWith("chatgpt-")); + updateData(id, DirpathRole, path); + updateData(id, FilesizeRole, toFileSize(info.size())); + } } } } @@ -470,9 +800,17 @@ void ModelList::updateModelsFromDirectory() processDirectory(exePath); if (localPath != exePath) processDirectory(localPath); + + if (installedModels()->count()) { + const QString firstModel = + installedModels()->firstId(); + QSettings settings; + settings.setValue("defaultModel", firstModel); + settings.sync(); + } } -void ModelList::updateModelList() +void ModelList::updateModelsFromJson() { #if defined(USE_LOCAL_MODELSJSON) QUrl jsonUrl("file://" + QDir::homePath() + "/dev/large_language_models/gpt4all/gpt4all-chat/metadata/models.json"); @@ -498,8 +836,9 @@ void ModelList::updateModelList() delete jsonReply; } -static bool operator==(const ModelInfo& lhs, const ModelInfo& rhs) { - return lhs.filename == rhs.filename && lhs.md5sum == rhs.md5sum; +void ModelList::updateDataForSettings() +{ + emit dataChanged(index(0, 0), index(m_models.size() - 1, 0)); } static bool compareVersions(const QString &a, const QString &b) { @@ -566,24 +905,45 @@ void ModelList::parseModelsJsonFile(const QByteArray &jsonData) modelFilesize = ModelList::toFileSize(modelFilesize.toULongLong()); - if (!contains(modelFilename)) - addModel(modelFilename); + const QString id = modelName; + Q_ASSERT(!id.isEmpty()); - if (!modelName.isEmpty()) - updateData(modelFilename, ModelList::NameRole, modelName); - updateData(modelFilename, ModelList::FilesizeRole, modelFilesize); - updateData(modelFilename, ModelList::Md5sumRole, modelMd5sum); - updateData(modelFilename, ModelList::DefaultRole, isDefault); - updateData(modelFilename, ModelList::DescriptionRole, description); - updateData(modelFilename, ModelList::RequiresVersionRole, requiresVersion); - updateData(modelFilename, ModelList::DeprecatedVersionRole, deprecatedVersion); - updateData(modelFilename, ModelList::UrlRole, url); - updateData(modelFilename, ModelList::DisableGUIRole, disableGUI); - updateData(modelFilename, ModelList::OrderRole, order); - updateData(modelFilename, ModelList::RamrequiredRole, ramrequired); - updateData(modelFilename, ModelList::ParametersRole, parameters); - updateData(modelFilename, ModelList::QuantRole, quant); - updateData(modelFilename, ModelList::TypeRole, type); + if (!contains(id)) + addModel(id); + + updateData(id, ModelList::NameRole, modelName); + updateData(id, ModelList::FilenameRole, modelFilename); + updateData(id, ModelList::FilesizeRole, modelFilesize); + updateData(id, ModelList::Md5sumRole, modelMd5sum); + updateData(id, ModelList::DefaultRole, isDefault); + updateData(id, ModelList::DescriptionRole, description); + updateData(id, ModelList::RequiresVersionRole, requiresVersion); + updateData(id, ModelList::DeprecatedVersionRole, deprecatedVersion); + updateData(id, ModelList::UrlRole, url); + updateData(id, ModelList::DisableGUIRole, disableGUI); + updateData(id, ModelList::OrderRole, order); + updateData(id, ModelList::RamrequiredRole, ramrequired); + updateData(id, ModelList::ParametersRole, parameters); + updateData(id, ModelList::QuantRole, quant); + updateData(id, ModelList::TypeRole, type); + if (obj.contains("temperature")) + updateData(id, ModelList::TemperatureRole, obj["temperature"].toDouble()); + if (obj.contains("topP")) + updateData(id, ModelList::TopPRole, obj["topP"].toDouble()); + if (obj.contains("topK")) + updateData(id, ModelList::TopKRole, obj["topK"].toInt()); + if (obj.contains("maxLength")) + updateData(id, ModelList::MaxLengthRole, obj["maxLength"].toInt()); + if (obj.contains("promptBatchSize")) + updateData(id, ModelList::PromptBatchSizeRole, obj["promptBatchSize"].toInt()); + if (obj.contains("repeatPenalty")) + updateData(id, ModelList::RepeatPenaltyRole, obj["repeatPenalty"].toDouble()); + if (obj.contains("repeatPenaltyTokens")) + updateData(id, ModelList::RepeatPenaltyTokensRole, obj["repeatPenaltyTokens"].toInt()); + if (obj.contains("promptTemplate")) + updateData(id, ModelList::PromptTemplateRole, obj["promptTemplate"].toString()); + if (obj.contains("systemPrompt")) + updateData(id, ModelList::SystemPromptRole, obj["systemPrompt"].toString()); } const QString chatGPTDesc = tr("
  • Requires personal OpenAI API key.
  • WARNING: Will send" @@ -592,44 +952,105 @@ void ModelList::parseModelsJsonFile(const QByteArray &jsonData) " here.
    • "); { + const QString modelName = "ChatGPT-3.5 Turbo"; + const QString id = modelName; const QString modelFilename = "chatgpt-gpt-3.5-turbo.txt"; - if (!contains(modelFilename)) - addModel(modelFilename); - updateData(modelFilename, ModelList::NameRole, "ChatGPT-3.5 Turbo"); - updateData(modelFilename, ModelList::FilesizeRole, "minimal"); - updateData(modelFilename, ModelList::ChatGPTRole, true); - updateData(modelFilename, ModelList::DescriptionRole, + if (!contains(id)) + addModel(id); + updateData(id, ModelList::NameRole, modelName); + updateData(id, ModelList::FilenameRole, modelFilename); + updateData(id, ModelList::FilesizeRole, "minimal"); + updateData(id, ModelList::ChatGPTRole, true); + updateData(id, ModelList::DescriptionRole, tr("OpenAI's ChatGPT model GPT-3.5 Turbo
    ") + chatGPTDesc); - updateData(modelFilename, ModelList::RequiresVersionRole, "2.4.2"); - updateData(modelFilename, ModelList::OrderRole, "ca"); - updateData(modelFilename, ModelList::RamrequiredRole, 0); - updateData(modelFilename, ModelList::ParametersRole, "?"); - updateData(modelFilename, ModelList::QuantRole, "NA"); - updateData(modelFilename, ModelList::TypeRole, "GPT"); + updateData(id, ModelList::RequiresVersionRole, "2.4.2"); + updateData(id, ModelList::OrderRole, "ca"); + updateData(id, ModelList::RamrequiredRole, 0); + updateData(id, ModelList::ParametersRole, "?"); + updateData(id, ModelList::QuantRole, "NA"); + updateData(id, ModelList::TypeRole, "GPT"); } { + const QString modelName = "ChatGPT-4"; + const QString id = modelName; const QString modelFilename = "chatgpt-gpt-4.txt"; - if (!contains(modelFilename)) - addModel(modelFilename); - updateData(modelFilename, ModelList::NameRole, "ChatGPT-4"); - updateData(modelFilename, ModelList::FilesizeRole, "minimal"); - updateData(modelFilename, ModelList::ChatGPTRole, true); - updateData(modelFilename, ModelList::DescriptionRole, + if (!contains(id)) + addModel(id); + updateData(id, ModelList::NameRole, modelName); + updateData(id, ModelList::FilenameRole, modelFilename); + updateData(id, ModelList::FilesizeRole, "minimal"); + updateData(id, ModelList::ChatGPTRole, true); + updateData(id, ModelList::DescriptionRole, tr("OpenAI's ChatGPT model GPT-4
    ") + chatGPTDesc); - updateData(modelFilename, ModelList::RequiresVersionRole, "2.4.2"); - updateData(modelFilename, ModelList::OrderRole, "cb"); - updateData(modelFilename, ModelList::RamrequiredRole, 0); - updateData(modelFilename, ModelList::ParametersRole, "?"); - updateData(modelFilename, ModelList::QuantRole, "NA"); - updateData(modelFilename, ModelList::TypeRole, "GPT"); + updateData(id, ModelList::RequiresVersionRole, "2.4.2"); + updateData(id, ModelList::OrderRole, "cb"); + updateData(id, ModelList::RamrequiredRole, 0); + updateData(id, ModelList::ParametersRole, "?"); + updateData(id, ModelList::QuantRole, "NA"); + updateData(id, ModelList::TypeRole, "GPT"); } if (installedModels()->count()) { const QString firstModel = - installedModels()->firstFilename(); + installedModels()->firstId(); QSettings settings; settings.setValue("defaultModel", firstModel); settings.sync(); } } + +void ModelList::updateModelsFromSettings() +{ + QSettings settings; + settings.sync(); + QStringList groups = settings.childGroups(); + for (const QString g : groups) { + if (!g.startsWith("model-")) + continue; + + const QString id = g.sliced(6); + if (contains(id)) + continue; + + if (!settings.contains(g+ "/isClone")) + continue; + + Q_ASSERT(settings.contains(g + "/name")); + const QString name = settings.value(g + "/name").toString(); + Q_ASSERT(settings.contains(g + "/filename")); + const QString filename = settings.value(g + "/filename").toString(); + Q_ASSERT(settings.contains(g + "/temperature")); + const double temperature = settings.value(g + "/temperature").toDouble(); + Q_ASSERT(settings.contains(g + "/topP")); + const double topP = settings.value(g + "/topP").toDouble(); + Q_ASSERT(settings.contains(g + "/topK")); + const int topK = settings.value(g + "/topK").toInt(); + Q_ASSERT(settings.contains(g + "/maxLength")); + const int maxLength = settings.value(g + "/maxLength").toInt(); + Q_ASSERT(settings.contains(g + "/promptBatchSize")); + const int promptBatchSize = settings.value(g + "/promptBatchSize").toInt(); + Q_ASSERT(settings.contains(g + "/repeatPenalty")); + const double repeatPenalty = settings.value(g + "/repeatPenalty").toDouble(); + Q_ASSERT(settings.contains(g + "/repeatPenaltyTokens")); + const int repeatPenaltyTokens = settings.value(g + "/repeatPenaltyTokens").toInt(); + Q_ASSERT(settings.contains(g + "/promptTemplate")); + const QString promptTemplate = settings.value(g + "/promptTemplate").toString(); + Q_ASSERT(settings.contains(g + "/systemPrompt")); + const QString systemPrompt = settings.value(g + "/systemPrompt").toString(); + + addModel(id); + updateData(id, ModelList::IsCloneRole, true); + updateData(id, ModelList::NameRole, name); + updateData(id, ModelList::FilenameRole, filename); + updateData(id, ModelList::TemperatureRole, temperature); + updateData(id, ModelList::TopPRole, topP); + updateData(id, ModelList::TopKRole, topK); + updateData(id, ModelList::MaxLengthRole, maxLength); + updateData(id, ModelList::PromptBatchSizeRole, promptBatchSize); + updateData(id, ModelList::RepeatPenaltyRole, repeatPenalty); + updateData(id, ModelList::RepeatPenaltyTokensRole, repeatPenaltyTokens); + updateData(id, ModelList::PromptTemplateRole, promptTemplate); + updateData(id, ModelList::SystemPromptRole, systemPrompt); + } +} diff --git a/gpt4all-chat/modellist.h b/gpt4all-chat/modellist.h index 7ad465e6..dc3452ce 100644 --- a/gpt4all-chat/modellist.h +++ b/gpt4all-chat/modellist.h @@ -6,8 +6,9 @@ struct ModelInfo { Q_GADGET - Q_PROPERTY(QString name MEMBER name) - Q_PROPERTY(QString filename MEMBER filename) + Q_PROPERTY(QString id READ id WRITE setId) + Q_PROPERTY(QString name READ name WRITE setName) + Q_PROPERTY(QString filename READ filename WRITE setFilename) Q_PROPERTY(QString dirpath MEMBER dirpath) Q_PROPERTY(QString filesize MEMBER filesize) Q_PROPERTY(QByteArray md5sum MEMBER md5sum) @@ -32,10 +33,27 @@ struct ModelInfo { Q_PROPERTY(QString parameters MEMBER parameters) Q_PROPERTY(QString quant MEMBER quant) Q_PROPERTY(QString type MEMBER type) + Q_PROPERTY(bool isClone MEMBER isClone) + Q_PROPERTY(double temperature READ temperature WRITE setTemperature) + Q_PROPERTY(double topP READ topP WRITE setTopP) + Q_PROPERTY(int topK READ topK WRITE setTopK) + Q_PROPERTY(int maxLength READ maxLength WRITE setMaxLength) + Q_PROPERTY(int promptBatchSize READ promptBatchSize WRITE setPromptBatchSize) + Q_PROPERTY(double repeatPenalty READ repeatPenalty WRITE setRepeatPenalty) + Q_PROPERTY(int repeatPenaltyTokens READ repeatPenaltyTokens WRITE setRepeatPenaltyTokens) + Q_PROPERTY(QString promptTemplate READ promptTemplate WRITE setPromptTemplate) + Q_PROPERTY(QString systemPrompt READ systemPrompt WRITE setSystemPrompt) public: - QString name; - QString filename; + QString id() const; + void setId(const QString &id); + + QString name() const; + void setName(const QString &name); + + QString filename() const; + void setFilename(const QString &name); + QString dirpath; QString filesize; QByteArray md5sum; @@ -60,9 +78,45 @@ public: QString parameters; QString quant; QString type; + bool isClone = false; + bool operator==(const ModelInfo &other) const { - return filename == other.filename; + return m_id == other.m_id; } + + double temperature() const; + void setTemperature(double t); + double topP() const; + void setTopP(double p); + int topK() const; + void setTopK(int k); + int maxLength() const; + void setMaxLength(int l); + int promptBatchSize() const; + void setPromptBatchSize(int s); + double repeatPenalty() const; + void setRepeatPenalty(double p); + int repeatPenaltyTokens() const; + void setRepeatPenaltyTokens(int t); + QString promptTemplate() const; + void setPromptTemplate(const QString &t); + QString systemPrompt() const; + void setSystemPrompt(const QString &p); + +private: + QString m_id; + QString m_name; + QString m_filename; + double m_temperature = 0.7; + double m_topP = 0.1; + int m_topK = 40; + int m_maxLength = 4096; + int m_promptBatchSize = 128; + double m_repeatPenalty = 1.18; + int m_repeatPenaltyTokens = 64; + QString m_promptTemplate = "### Human:\n%1\n### Assistant:\n"; + QString m_systemPrompt = "### System:\nYou are an AI assistant who gives quality response to whatever humans ask of you.\n"; + friend class MySettings; }; Q_DECLARE_METATYPE(ModelInfo) @@ -73,7 +127,7 @@ class InstalledModels : public QSortFilterProxyModel public: explicit InstalledModels(QObject *parent); int count() const; - QString firstFilename() const; + QString firstId() const; Q_SIGNALS: void countChanged(); @@ -120,7 +174,8 @@ public: static ModelList *globalInstance(); enum Roles { - NameRole = Qt::UserRole + 1, + IdRole = Qt::UserRole + 1, + NameRole, FilenameRole, DirpathRole, FilesizeRole, @@ -145,12 +200,23 @@ public: RamrequiredRole, ParametersRole, QuantRole, - TypeRole + TypeRole, + IsCloneRole, + TemperatureRole, + TopPRole, + TopKRole, + MaxLengthRole, + PromptBatchSizeRole, + RepeatPenaltyRole, + RepeatPenaltyTokensRole, + PromptTemplateRole, + SystemPromptRole, }; QHash roleNames() const override { QHash roles; + roles[IdRole] = "id"; roles[NameRole] = "name"; roles[FilenameRole] = "filename"; roles[DirpathRole] = "dirpath"; @@ -177,21 +243,36 @@ public: roles[ParametersRole] = "parameters"; roles[QuantRole] = "quant"; roles[TypeRole] = "type"; + roles[IsCloneRole] = "isClone"; + roles[TemperatureRole] = "temperature"; + roles[TopPRole] = "topP"; + roles[TopKRole] = "topK"; + roles[MaxLengthRole] = "maxLength"; + roles[PromptBatchSizeRole] = "promptBatchSize"; + roles[RepeatPenaltyRole] = "repeatPenalty"; + roles[RepeatPenaltyTokensRole] = "repeatPenaltyTokens"; + roles[PromptTemplateRole] = "promptTemplate"; + roles[SystemPromptRole] = "systemPrompt"; return roles; } int rowCount(const QModelIndex &parent = QModelIndex()) const override; QVariant data(const QModelIndex &index, int role = Qt::DisplayRole) const override; - QVariant data(const QString &filename, int role) const; - void updateData(const QString &filename, int role, const QVariant &value); + QVariant data(const QString &id, int role) const; + void updateData(const QString &id, int role, const QVariant &value); int count() const { return m_models.size(); } - bool contains(const QString &filename) const; - Q_INVOKABLE ModelInfo modelInfo(const QString &filename) const; + bool contains(const QString &id) const; + bool containsByFilename(const QString &filename) const; + Q_INVOKABLE ModelInfo modelInfo(const QString &id) const; + Q_INVOKABLE ModelInfo modelInfoByFilename(const QString &filename) const; + Q_INVOKABLE bool isUniqueName(const QString &name) const; + Q_INVOKABLE QString clone(const ModelInfo &model); + Q_INVOKABLE void remove(const ModelInfo &model); ModelInfo defaultModelInfo() const; - void addModel(const QString &filename); + void addModel(const QString &id); const QList exportModelList() const; const QList userDefaultModelList() const; @@ -220,8 +301,10 @@ Q_SIGNALS: void userDefaultModelListChanged(); private Q_SLOTS: + void updateModelsFromJson(); + void updateModelsFromSettings(); void updateModelsFromDirectory(); - void updateModelList(); + void updateDataForSettings(); private: QString modelDirPath(const QString &modelName, bool isChatGPT); @@ -229,6 +312,7 @@ private: QVariant dataInternal(const ModelInfo *info, int role) const; static bool lessThan(const ModelInfo* a, const ModelInfo* b); void parseModelsJsonFile(const QByteArray &jsonData); + QString uniqueModelName(const ModelInfo &model) const; private: mutable QMutex m_mutex; diff --git a/gpt4all-chat/mysettings.cpp b/gpt4all-chat/mysettings.cpp index f8070a65..e2e3097c 100644 --- a/gpt4all-chat/mysettings.cpp +++ b/gpt4all-chat/mysettings.cpp @@ -1,4 +1,5 @@ #include "mysettings.h" +#include "modellist.h" #include #include @@ -7,14 +8,6 @@ #include #include -static double default_temperature = 0.7; -static double default_topP = 0.1; -static int default_topK = 40; -static int default_maxLength = 4096; -static int default_promptBatchSize = 128; -static double default_repeatPenalty = 1.18; -static int default_repeatPenaltyTokens = 64; -static QString default_promptTemplate = "### Human:\n%1\n### Assistant:\n"; static int default_threadCount = 0; static bool default_saveChats = false; static bool default_saveChatGPTChats = true; @@ -68,16 +61,17 @@ MySettings::MySettings() QSettings::setDefaultFormat(QSettings::IniFormat); } -void MySettings::restoreGenerationDefaults() +void MySettings::restoreModelDefaults(const ModelInfo &model) { - setTemperature(default_temperature); - setTopP(default_topP); - setTopK(default_topK); - setMaxLength(default_maxLength); - setPromptBatchSize(default_promptBatchSize); - setRepeatPenalty(default_repeatPenalty); - setRepeatPenaltyTokens(default_repeatPenaltyTokens); - setPromptTemplate(default_promptTemplate); + setModelTemperature(model, model.m_temperature); + setModelTopP(model, model.m_topP); + setModelTopK(model, model.m_topK);; + setModelMaxLength(model, model.m_maxLength); + setModelPromptBatchSize(model, model.m_promptBatchSize); + setModelRepeatPenalty(model, model.m_repeatPenalty); + setModelRepeatPenaltyTokens(model, model.m_repeatPenaltyTokens); + setModelPromptTemplate(model, model.m_promptTemplate); + setModelSystemPrompt(model, model.m_systemPrompt); } void MySettings::restoreApplicationDefaults() @@ -97,148 +91,256 @@ void MySettings::restoreLocalDocsDefaults() setLocalDocsRetrievalSize(default_localDocsRetrievalSize); } -double MySettings::temperature() const +void MySettings::eraseModel(const ModelInfo &m) +{ + QSettings settings; + settings.remove(QString("model-%1").arg(m.id())); + settings.sync(); +} + +QString MySettings::modelName(const ModelInfo &m) const { QSettings setting; setting.sync(); - return setting.value("temperature", default_temperature).toDouble(); + return setting.value(QString("model-%1").arg(m.id()) + "/name", + !m.m_name.isEmpty() ? m.m_name : m.m_filename).toString(); } -void MySettings::setTemperature(double t) +void MySettings::setModelName(const ModelInfo &m, const QString &name, bool force) { - if (temperature() == t) + if ((modelName(m) == name || m.id().isEmpty()) && !force) return; QSettings setting; - setting.setValue("temperature", t); + if ((m.m_name == name || m.m_filename == name) && !m.isClone) + setting.remove(QString("model-%1").arg(m.id()) + "/name"); + else + setting.setValue(QString("model-%1").arg(m.id()) + "/name", name); + if (m.isClone) + setting.setValue(QString("model-%1").arg(m.id()) + "/isClone", "true"); setting.sync(); - emit temperatureChanged(); + if (!force) + emit nameChanged(m); } -double MySettings::topP() const +QString MySettings::modelFilename(const ModelInfo &m) const { QSettings setting; setting.sync(); - return setting.value("topP", default_topP).toDouble(); + return setting.value(QString("model-%1").arg(m.id()) + "/filename", m.m_filename).toString(); } -void MySettings::setTopP(double p) +void MySettings::setModelFilename(const ModelInfo &m, const QString &filename, bool force) { - if (topP() == p) + if ((modelFilename(m) == filename || m.id().isEmpty()) && !force) return; QSettings setting; - setting.setValue("topP", p); + if (m.m_filename == filename && !m.isClone) + setting.remove(QString("model-%1").arg(m.id()) + "/filename"); + else + setting.setValue(QString("model-%1").arg(m.id()) + "/filename", filename); setting.sync(); - emit topPChanged(); + if (!force) + emit filenameChanged(m); } -int MySettings::topK() const +double MySettings::modelTemperature(const ModelInfo &m) const { QSettings setting; setting.sync(); - return setting.value("topK", default_topK).toInt(); + return setting.value(QString("model-%1").arg(m.id()) + "/temperature", m.m_temperature).toDouble(); } -void MySettings::setTopK(int k) +void MySettings::setModelTemperature(const ModelInfo &m, double t, bool force) { - if (topK() == k) + if (modelTemperature(m) == t && !force) return; QSettings setting; - setting.setValue("topK", k); + if (m.m_temperature == t && !m.isClone) + setting.remove(QString("model-%1").arg(m.id()) + "/temperature"); + else + setting.setValue(QString("model-%1").arg(m.id()) + "/temperature", t); setting.sync(); - emit topKChanged(); + if (!force) + emit temperatureChanged(m); } -int MySettings::maxLength() const +double MySettings::modelTopP(const ModelInfo &m) const { QSettings setting; setting.sync(); - return setting.value("maxLength", default_maxLength).toInt(); + return setting.value(QString("model-%1").arg(m.id()) + "/topP", m.m_topP).toDouble(); } -void MySettings::setMaxLength(int l) +void MySettings::setModelTopP(const ModelInfo &m, double p, bool force) { - if (maxLength() == l) + if (modelTopP(m) == p && !force) return; QSettings setting; - setting.setValue("maxLength", l); + if (m.m_topP == p && !m.isClone) + setting.remove(QString("model-%1").arg(m.id()) + "/topP"); + else + setting.setValue(QString("model-%1").arg(m.id()) + "/topP", p); setting.sync(); - emit maxLengthChanged(); + if (!force) + emit topPChanged(m); } -int MySettings::promptBatchSize() const +int MySettings::modelTopK(const ModelInfo &m) const { QSettings setting; setting.sync(); - return setting.value("promptBatchSize", default_promptBatchSize).toInt(); + return setting.value(QString("model-%1").arg(m.id()) + "/topK", m.m_topK).toInt(); } -void MySettings::setPromptBatchSize(int s) +void MySettings::setModelTopK(const ModelInfo &m, int k, bool force) { - if (promptBatchSize() == s) + if (modelTopK(m) == k && !force) return; QSettings setting; - setting.setValue("promptBatchSize", s); + if (m.m_topK == k && !m.isClone) + setting.remove(QString("model-%1").arg(m.id()) + "/topK"); + else + setting.setValue(QString("model-%1").arg(m.id()) + "/topK", k); setting.sync(); - emit promptBatchSizeChanged(); + if (!force) + emit topKChanged(m); } -double MySettings::repeatPenalty() const +int MySettings::modelMaxLength(const ModelInfo &m) const { QSettings setting; setting.sync(); - return setting.value("repeatPenalty", default_repeatPenalty).toDouble(); + return setting.value(QString("model-%1").arg(m.id()) + "/maxLength", m.m_maxLength).toInt(); } -void MySettings::setRepeatPenalty(double p) +void MySettings::setModelMaxLength(const ModelInfo &m, int l, bool force) { - if (repeatPenalty() == p) + if (modelMaxLength(m) == l && !force) return; QSettings setting; - setting.setValue("repeatPenalty", p); + if (m.m_maxLength == l && !m.isClone) + setting.remove(QString("model-%1").arg(m.id()) + "/maxLength"); + else + setting.setValue(QString("model-%1").arg(m.id()) + "/maxLength", l); setting.sync(); - emit repeatPenaltyChanged(); + if (!force) + emit maxLengthChanged(m); } -int MySettings::repeatPenaltyTokens() const +int MySettings::modelPromptBatchSize(const ModelInfo &m) const { QSettings setting; setting.sync(); - return setting.value("repeatPenaltyTokens", default_repeatPenaltyTokens).toInt(); + return setting.value(QString("model-%1").arg(m.id()) + "/promptBatchSize", m.m_promptBatchSize).toInt(); } -void MySettings::setRepeatPenaltyTokens(int t) +void MySettings::setModelPromptBatchSize(const ModelInfo &m, int s, bool force) { - if (repeatPenaltyTokens() == t) + if (modelPromptBatchSize(m) == s && !force) return; QSettings setting; - setting.setValue("repeatPenaltyTokens", t); + if (m.m_promptBatchSize == s && !m.isClone) + setting.remove(QString("model-%1").arg(m.id()) + "/promptBatchSize"); + else + setting.setValue(QString("model-%1").arg(m.id()) + "/promptBatchSize", s); setting.sync(); - emit repeatPenaltyTokensChanged(); + if (!force) + emit promptBatchSizeChanged(m); } -QString MySettings::promptTemplate() const +double MySettings::modelRepeatPenalty(const ModelInfo &m) const { QSettings setting; setting.sync(); - return setting.value("promptTemplate", default_promptTemplate).toString(); + return setting.value(QString("model-%1").arg(m.id()) + "/repeatPenalty", m.m_repeatPenalty).toDouble(); } -void MySettings::setPromptTemplate(const QString &t) +void MySettings::setModelRepeatPenalty(const ModelInfo &m, double p, bool force) { - if (promptTemplate() == t) + if (modelRepeatPenalty(m) == p && !force) return; QSettings setting; - setting.setValue("promptTemplate", t); + if (m.m_repeatPenalty == p && !m.isClone) + setting.remove(QString("model-%1").arg(m.id()) + "/repeatPenalty"); + else + setting.setValue(QString("model-%1").arg(m.id()) + "/repeatPenalty", p); setting.sync(); - emit promptTemplateChanged(); + if (!force) + emit repeatPenaltyChanged(m); +} + +int MySettings::modelRepeatPenaltyTokens(const ModelInfo &m) const +{ + QSettings setting; + setting.sync(); + return setting.value(QString("model-%1").arg(m.id()) + "/repeatPenaltyTokens", m.m_repeatPenaltyTokens).toInt(); +} + +void MySettings::setModelRepeatPenaltyTokens(const ModelInfo &m, int t, bool force) +{ + if (modelRepeatPenaltyTokens(m) == t && !force) + return; + + QSettings setting; + if (m.m_repeatPenaltyTokens == t && !m.isClone) + setting.remove(QString("model-%1").arg(m.id()) + "/repeatPenaltyTokens"); + else + setting.setValue(QString("model-%1").arg(m.id()) + "/repeatPenaltyTokens", t); + setting.sync(); + if (!force) + emit repeatPenaltyTokensChanged(m); +} + +QString MySettings::modelPromptTemplate(const ModelInfo &m) const +{ + QSettings setting; + setting.sync(); + return setting.value(QString("model-%1").arg(m.id()) + "/promptTemplate", m.m_promptTemplate).toString(); +} + +void MySettings::setModelPromptTemplate(const ModelInfo &m, const QString &t, bool force) +{ + if (modelPromptTemplate(m) == t && !force) + return; + + QSettings setting; + if (m.m_promptTemplate == t && !m.isClone) + setting.remove(QString("model-%1").arg(m.id()) + "/promptTemplate"); + else + setting.setValue(QString("model-%1").arg(m.id()) + "/promptTemplate", t); + setting.sync(); + if (!force) + emit promptTemplateChanged(m); +} + +QString MySettings::modelSystemPrompt(const ModelInfo &m) const +{ + QSettings setting; + setting.sync(); + return setting.value(QString("model-%1").arg(m.id()) + "/systemPrompt", m.m_systemPrompt).toString(); +} + +void MySettings::setModelSystemPrompt(const ModelInfo &m, const QString &p, bool force) +{ + if (modelSystemPrompt(m) == p && !force) + return; + + QSettings setting; + if (m.m_systemPrompt == p && !m.isClone) + setting.remove(QString("model-%1").arg(m.id()) + "/systemPrompt"); + else + setting.setValue(QString("model-%1").arg(m.id()) + "/systemPrompt", p); + setting.sync(); + if (!force) + emit systemPromptChanged(m); } int MySettings::threadCount() const diff --git a/gpt4all-chat/mysettings.h b/gpt4all-chat/mysettings.h index 149350c6..4688ceed 100644 --- a/gpt4all-chat/mysettings.h +++ b/gpt4all-chat/mysettings.h @@ -4,17 +4,10 @@ #include #include +class ModelInfo; class MySettings : public QObject { Q_OBJECT - Q_PROPERTY(double temperature READ temperature WRITE setTemperature NOTIFY temperatureChanged) - Q_PROPERTY(double topP READ topP WRITE setTopP NOTIFY topPChanged) - Q_PROPERTY(int topK READ topK WRITE setTopK NOTIFY topKChanged) - Q_PROPERTY(int maxLength READ maxLength WRITE setMaxLength NOTIFY maxLengthChanged) - Q_PROPERTY(int promptBatchSize READ promptBatchSize WRITE setPromptBatchSize NOTIFY promptBatchSizeChanged) - Q_PROPERTY(double repeatPenalty READ repeatPenalty WRITE setRepeatPenalty NOTIFY repeatPenaltyChanged) - Q_PROPERTY(int repeatPenaltyTokens READ repeatPenaltyTokens WRITE setRepeatPenaltyTokens NOTIFY repeatPenaltyTokensChanged) - Q_PROPERTY(QString promptTemplate READ promptTemplate WRITE setPromptTemplate NOTIFY promptTemplateChanged) Q_PROPERTY(int threadCount READ threadCount WRITE setThreadCount NOTIFY threadCountChanged) Q_PROPERTY(bool saveChats READ saveChats WRITE setSaveChats NOTIFY saveChatsChanged) Q_PROPERTY(bool saveChatGPTChats READ saveChatGPTChats WRITE setSaveChatGPTChats NOTIFY saveChatGPTChatsChanged) @@ -33,27 +26,34 @@ public: static MySettings *globalInstance(); // Restore methods - Q_INVOKABLE void restoreGenerationDefaults(); + Q_INVOKABLE void restoreModelDefaults(const ModelInfo &model); Q_INVOKABLE void restoreApplicationDefaults(); Q_INVOKABLE void restoreLocalDocsDefaults(); - // Generation settings - double temperature() const; - void setTemperature(double t); - double topP() const; - void setTopP(double p); - int topK() const; - void setTopK(int k); - int maxLength() const; - void setMaxLength(int l); - int promptBatchSize() const; - void setPromptBatchSize(int s); - double repeatPenalty() const; - void setRepeatPenalty(double p); - int repeatPenaltyTokens() const; - void setRepeatPenaltyTokens(int t); - QString promptTemplate() const; - void setPromptTemplate(const QString &t); + // Model/Character settings + void eraseModel(const ModelInfo &m); + QString modelName(const ModelInfo &m) const; + Q_INVOKABLE void setModelName(const ModelInfo &m, const QString &name, bool force = false); + QString modelFilename(const ModelInfo &m) const; + Q_INVOKABLE void setModelFilename(const ModelInfo &m, const QString &filename, bool force = false); + double modelTemperature(const ModelInfo &m) const; + Q_INVOKABLE void setModelTemperature(const ModelInfo &m, double t, bool force = false); + double modelTopP(const ModelInfo &m) const; + Q_INVOKABLE void setModelTopP(const ModelInfo &m, double p, bool force = false); + int modelTopK(const ModelInfo &m) const; + Q_INVOKABLE void setModelTopK(const ModelInfo &m, int k, bool force = false); + int modelMaxLength(const ModelInfo &m) const; + Q_INVOKABLE void setModelMaxLength(const ModelInfo &m, int l, bool force = false); + int modelPromptBatchSize(const ModelInfo &m) const; + Q_INVOKABLE void setModelPromptBatchSize(const ModelInfo &m, int s, bool force = false); + double modelRepeatPenalty(const ModelInfo &m) const; + Q_INVOKABLE void setModelRepeatPenalty(const ModelInfo &m, double p, bool force = false); + int modelRepeatPenaltyTokens(const ModelInfo &m) const; + Q_INVOKABLE void setModelRepeatPenaltyTokens(const ModelInfo &m, int t, bool force = false); + QString modelPromptTemplate(const ModelInfo &m) const; + Q_INVOKABLE void setModelPromptTemplate(const ModelInfo &m, const QString &t, bool force = false); + QString modelSystemPrompt(const ModelInfo &m) const; + Q_INVOKABLE void setModelSystemPrompt(const ModelInfo &m, const QString &p, bool force = false); // Application settings int threadCount() const; @@ -90,14 +90,17 @@ public: void setNetworkUsageStatsActive(bool b); Q_SIGNALS: - void temperatureChanged(); - void topPChanged(); - void topKChanged(); - void maxLengthChanged(); - void promptBatchSizeChanged(); - void repeatPenaltyChanged(); - void repeatPenaltyTokensChanged(); - void promptTemplateChanged(); + void nameChanged(const ModelInfo &model); + void filenameChanged(const ModelInfo &model); + void temperatureChanged(const ModelInfo &model); + void topPChanged(const ModelInfo &model); + void topKChanged(const ModelInfo &model); + void maxLengthChanged(const ModelInfo &model); + void promptBatchSizeChanged(const ModelInfo &model); + void repeatPenaltyChanged(const ModelInfo &model); + void repeatPenaltyTokensChanged(const ModelInfo &model); + void promptTemplateChanged(const ModelInfo &model); + void systemPromptChanged(const ModelInfo &model); void threadCountChanged(); void saveChatsChanged(); void saveChatGPTChatsChanged(); diff --git a/gpt4all-chat/network.cpp b/gpt4all-chat/network.cpp index e69b715c..61ebfa61 100644 --- a/gpt4all-chat/network.cpp +++ b/gpt4all-chat/network.cpp @@ -88,7 +88,7 @@ bool Network::packageAndSendJson(const QString &ingestId, const QString &json) Q_ASSERT(ChatListModel::globalInstance()->currentChat()); QJsonObject object = doc.object(); object.insert("source", "gpt4all-chat"); - object.insert("agent_id", ChatListModel::globalInstance()->currentChat()->modelInfo().filename); + object.insert("agent_id", ChatListModel::globalInstance()->currentChat()->modelInfo().filename()); object.insert("submitter_id", m_uniqueId); object.insert("ingest_id", ingestId); @@ -96,7 +96,7 @@ bool Network::packageAndSendJson(const QString &ingestId, const QString &json) if (!attribution.isEmpty()) object.insert("network/attribution", attribution); - QString promptTemplate = MySettings::globalInstance()->promptTemplate(); + QString promptTemplate = ChatListModel::globalInstance()->currentChat()->modelInfo().promptTemplate(); object.insert("prompt_template", promptTemplate); QJsonDocument newDoc; @@ -391,7 +391,7 @@ void Network::sendMixpanelEvent(const QString &ev, const QVector &valu properties.insert("ip", m_ipify); properties.insert("name", QCoreApplication::applicationName() + " v" + QCoreApplication::applicationVersion()); - properties.insert("model", ChatListModel::globalInstance()->currentChat()->modelInfo().filename); + properties.insert("model", ChatListModel::globalInstance()->currentChat()->modelInfo().filename()); // Some additional startup information if (ev == "startup") { diff --git a/gpt4all-chat/qml/ApplicationSettings.qml b/gpt4all-chat/qml/ApplicationSettings.qml index 85379488..4e475ac3 100644 --- a/gpt4all-chat/qml/ApplicationSettings.qml +++ b/gpt4all-chat/qml/ApplicationSettings.qml @@ -6,8 +6,12 @@ import QtQuick.Layouts import QtQuick.Dialogs import modellist import mysettings +import network MySettingsTab { + onRestoreDefaultsClicked: { + MySettings.restoreApplicationDefaults(); + } title: qsTr("Application") contentItem: GridLayout { id: applicationSettingsTabInner @@ -26,7 +30,9 @@ MySettingsTab { id: comboBox Layout.row: 1 Layout.column: 1 + Layout.columnSpan: 2 Layout.minimumWidth: 350 + Layout.fillWidth: true model: ModelList.userDefaultModelList Accessible.role: Accessible.ComboBox Accessible.name: qsTr("ComboBox for displaying/picking the default model") @@ -178,25 +184,30 @@ MySettingsTab { Layout.columnSpan: 3 Layout.fillWidth: true height: 1 - color: theme.dialogBorder + color: theme.tabBorder } + } + advancedSettings: GridLayout { + columns: 3 + rowSpacing: 10 + columnSpacing: 10 Rectangle { - Layout.row: 9 + Layout.row: 2 Layout.column: 0 Layout.fillWidth: true Layout.columnSpan: 3 height: 1 - color: theme.dialogBorder + color: theme.tabBorder } Label { id: gpuOverrideLabel text: qsTr("Force Metal (macOS+arm):") color: theme.textColor - Layout.row: 8 + Layout.row: 1 Layout.column: 0 } RowLayout { - Layout.row: 8 + Layout.row: 1 Layout.column: 1 Layout.columnSpan: 2 MyCheckBox { @@ -206,24 +217,18 @@ MySettingsTab { MySettings.forceMetal = !MySettings.forceMetal } } - Label { - id: warningLabel - Layout.maximumWidth: 730 + + Item { + Layout.fillWidth: true Layout.alignment: Qt.AlignTop - color: theme.textErrorColor - wrapMode: Text.WordWrap - text: qsTr("WARNING: On macOS with arm (M1+) this setting forces usage of the GPU. Can cause crashes if the model requires more RAM than the system supports. Because of crash possibility the setting will not persist across restarts of the application. This has no effect on non-macs or intel.") - } - } - MyButton { - Layout.row: 10 - Layout.column: 1 - Layout.columnSpan: 2 - Layout.fillWidth: true - text: qsTr("Restore Defaults") - Accessible.description: qsTr("Restores the settings dialog to a default state") - onClicked: { - MySettings.restoreApplicationDefaults(); + Layout.minimumHeight: warningLabel.height + Label { + id: warningLabel + width: parent.width + color: theme.textErrorColor + wrapMode: Text.WordWrap + text: qsTr("WARNING: On macOS with arm (M1+) this setting forces usage of the GPU. Can cause crashes if the model requires more RAM than the system supports. Because of crash possibility the setting will not persist across restarts of the application. This has no effect on non-macs or intel.") + } } } } diff --git a/gpt4all-chat/qml/GenerationSettings.qml b/gpt4all-chat/qml/GenerationSettings.qml deleted file mode 100644 index 6f14dfbc..00000000 --- a/gpt4all-chat/qml/GenerationSettings.qml +++ /dev/null @@ -1,302 +0,0 @@ -import QtCore -import QtQuick -import QtQuick.Controls -import QtQuick.Controls.Basic -import QtQuick.Layouts -import localdocs -import mysettings - -MySettingsTab { - title: qsTr("Presets") - contentItem: GridLayout { - id: generationSettingsTabInner - columns: 2 - rowSpacing: 10 - columnSpacing: 10 - - Label { - id: tempLabel - text: qsTr("Temperature:") - color: theme.textColor - Layout.row: 0 - Layout.column: 0 - } - MyTextField { - text: MySettings.temperature - color: theme.textColor - ToolTip.text: qsTr("Temperature increases the chances of choosing less likely tokens.\nNOTE: Higher temperature gives more creative but less predictable outputs.") - ToolTip.visible: hovered - Layout.row: 0 - Layout.column: 1 - validator: DoubleValidator { - locale: "C" - } - onEditingFinished: { - var val = parseFloat(text) - if (!isNaN(val)) { - MySettings.temperature = val - focus = false - } else { - text = MySettings.temperature - } - } - Accessible.role: Accessible.EditableText - Accessible.name: tempLabel.text - Accessible.description: ToolTip.text - } - Label { - id: topPLabel - text: qsTr("Top P:") - color: theme.textColor - Layout.row: 1 - Layout.column: 0 - } - MyTextField { - text: MySettings.topP - color: theme.textColor - ToolTip.text: qsTr("Only the most likely tokens up to a total probability of top_p can be chosen.\nNOTE: Prevents choosing highly unlikely tokens, aka Nucleus Sampling") - ToolTip.visible: hovered - Layout.row: 1 - Layout.column: 1 - validator: DoubleValidator { - locale: "C" - } - onEditingFinished: { - var val = parseFloat(text) - if (!isNaN(val)) { - MySettings.topP = val - focus = false - } else { - text = MySettings.topP - } - } - Accessible.role: Accessible.EditableText - Accessible.name: topPLabel.text - Accessible.description: ToolTip.text - } - Label { - id: topKLabel - text: qsTr("Top K:") - color: theme.textColor - Layout.row: 2 - Layout.column: 0 - } - MyTextField { - text: MySettings.topK - color: theme.textColor - ToolTip.text: qsTr("Only the top K most likely tokens will be chosen from") - ToolTip.visible: hovered - Layout.row: 2 - Layout.column: 1 - validator: IntValidator { - bottom: 1 - } - onEditingFinished: { - var val = parseInt(text) - if (!isNaN(val)) { - MySettings.topK = val - focus = false - } else { - text = MySettings.topK - } - } - Accessible.role: Accessible.EditableText - Accessible.name: topKLabel.text - Accessible.description: ToolTip.text - } - Label { - id: maxLengthLabel - text: qsTr("Max Length:") - color: theme.textColor - Layout.row: 3 - Layout.column: 0 - } - MyTextField { - text: MySettings.maxLength - color: theme.textColor - ToolTip.text: qsTr("Maximum length of response in tokens") - ToolTip.visible: hovered - Layout.row: 3 - Layout.column: 1 - validator: IntValidator { - bottom: 1 - } - onEditingFinished: { - var val = parseInt(text) - if (!isNaN(val)) { - MySettings.maxLength = val - focus = false - } else { - text = MySettings.maxLength - } - } - Accessible.role: Accessible.EditableText - Accessible.name: maxLengthLabel.text - Accessible.description: ToolTip.text - } - - Label { - id: batchSizeLabel - text: qsTr("Prompt Batch Size:") - color: theme.textColor - Layout.row: 4 - Layout.column: 0 - } - MyTextField { - text: MySettings.promptBatchSize - color: theme.textColor - ToolTip.text: qsTr("Amount of prompt tokens to process at once.\nNOTE: Higher values can speed up reading prompts but will use more RAM") - ToolTip.visible: hovered - Layout.row: 4 - Layout.column: 1 - validator: IntValidator { - bottom: 1 - } - onEditingFinished: { - var val = parseInt(text) - if (!isNaN(val)) { - MySettings.promptBatchSize = val - focus = false - } else { - text = MySettings.promptBatchSize - } - } - Accessible.role: Accessible.EditableText - Accessible.name: batchSizeLabel.text - Accessible.description: ToolTip.text - } - Label { - id: repeatPenaltyLabel - text: qsTr("Repeat Penalty:") - color: theme.textColor - Layout.row: 5 - Layout.column: 0 - } - MyTextField { - text: MySettings.repeatPenalty - color: theme.textColor - ToolTip.text: qsTr("Amount to penalize repetitiveness of the output") - ToolTip.visible: hovered - Layout.row: 5 - Layout.column: 1 - validator: DoubleValidator { - locale: "C" - } - onEditingFinished: { - var val = parseFloat(text) - if (!isNaN(val)) { - MySettings.repeatPenalty = val - focus = false - } else { - text = MySettings.repeatPenalty - } - } - Accessible.role: Accessible.EditableText - Accessible.name: repeatPenaltyLabel.text - Accessible.description: ToolTip.text - } - Label { - id: repeatPenaltyTokensLabel - text: qsTr("Repeat Penalty Tokens:") - color: theme.textColor - Layout.row: 6 - Layout.column: 0 - } - MyTextField { - text: MySettings.repeatPenaltyTokens - color: theme.textColor - ToolTip.text: qsTr("How far back in output to apply repeat penalty") - ToolTip.visible: hovered - Layout.row: 6 - Layout.column: 1 - validator: IntValidator { - bottom: 1 - } - onEditingFinished: { - var val = parseInt(text) - if (!isNaN(val)) { - MySettings.repeatPenaltyTokens = val - focus = false - } else { - text = MySettings.repeatPenaltyTokens - } - } - Accessible.role: Accessible.EditableText - Accessible.name: repeatPenaltyTokensLabel.text - Accessible.description: ToolTip.text - } - - ColumnLayout { - Layout.row: 7 - Layout.column: 0 - Layout.topMargin: 10 - Layout.alignment: Qt.AlignTop - spacing: 20 - - Label { - id: promptTemplateLabel - text: qsTr("Prompt Template:") - color: theme.textColor - } - - Label { - id: promptTemplateLabelHelp - Layout.maximumWidth: promptTemplateLabel.width - visible: templateTextArea.text.indexOf( - "%1") === -1 - color: theme.textErrorColor - text: qsTr("Must contain the string \"%1\" to be replaced with the user's input.") - wrapMode: TextArea.Wrap - Accessible.role: Accessible.EditableText - Accessible.name: text - } - } - - Rectangle { - Layout.row: 7 - Layout.column: 1 - Layout.fillWidth: true - height: 200 - color: "transparent" - clip: true - ScrollView { - id: templateScrollView - anchors.fill: parent - TextArea { - id: templateTextArea - text: MySettings.promptTemplate - color: theme.textColor - background: Rectangle { - implicitWidth: 150 - color: theme.backgroundLighter - radius: 10 - } - padding: 10 - wrapMode: TextArea.Wrap - onTextChanged: { - if (templateTextArea.text.indexOf("%1") !== -1) { - MySettings.promptTemplate = text - } - } - bottomPadding: 10 - Accessible.role: Accessible.EditableText - Accessible.name: promptTemplateLabel.text - Accessible.description: promptTemplateLabelHelp.text - ToolTip.text: qsTr("The prompt template partially determines how models will respond to prompts.\nNOTE: A longer, detailed template can lead to higher quality answers, but can also slow down generation.") - ToolTip.visible: hovered - } - } - } - MyButton { - Layout.row: 8 - Layout.column: 1 - Layout.fillWidth: true - text: qsTr("Restore Defaults") - Accessible.description: qsTr("Restores the settings dialog to a default state") - onClicked: { - MySettings.restoreGenerationDefaults(); - templateTextArea.text = MySettings.promptTemplate - } - } - } -} \ No newline at end of file diff --git a/gpt4all-chat/qml/LocalDocsSettings.qml b/gpt4all-chat/qml/LocalDocsSettings.qml index bafd191a..4bfc6e46 100644 --- a/gpt4all-chat/qml/LocalDocsSettings.qml +++ b/gpt4all-chat/qml/LocalDocsSettings.qml @@ -6,8 +6,13 @@ import QtQuick.Layouts import QtQuick.Dialogs import localdocs import mysettings +import network MySettingsTab { + onRestoreDefaultsClicked: { + MySettings.restoreLocalDocsDefaults(); + } + title: qsTr("LocalDocs Plugin (BETA)") contentItem: ColumnLayout { id: root @@ -25,101 +30,95 @@ MySettingsTab { } } - RowLayout { + Item { Layout.fillWidth: true - height: collection.height + 20 - spacing: 10 - MyTextField { - id: collection - width: 225 - horizontalAlignment: Text.AlignJustify - color: theme.textColor - placeholderText: qsTr("Collection name...") - placeholderTextColor: theme.mutedTextColor - ToolTip.text: qsTr("Name of the collection to add (Required)") - ToolTip.visible: hovered - Accessible.role: Accessible.EditableText - Accessible.name: collection.text - Accessible.description: ToolTip.text - function showError() { - collection.placeholderTextColor = theme.textErrorColor - } - onTextChanged: { - collection.placeholderTextColor = theme.mutedTextColor - } - } - - MyDirectoryField { - id: folderEdit - Layout.fillWidth: true - text: root.folder_path - placeholderText: qsTr("Folder path...") - placeholderTextColor: theme.mutedTextColor - ToolTip.text: qsTr("Folder path to documents (Required)") - ToolTip.visible: hovered - function showError() { - folderEdit.placeholderTextColor = theme.textErrorColor - } - onTextChanged: { - folderEdit.placeholderTextColor = theme.mutedTextColor - } - } - - MyButton { - id: browseButton - text: qsTr("Browse") - onClicked: { - folderDialog.open(); - } - } - - MyButton { - id: addButton - text: qsTr("Add") - Accessible.role: Accessible.Button - Accessible.name: text - Accessible.description: qsTr("Add button") - onClicked: { - var isError = false; - if (root.collection === "") { - isError = true; - collection.showError(); + height: row.height + RowLayout { + id: row + anchors.left: parent.left + anchors.right: parent.right + height: collection.height + spacing: 10 + MyTextField { + id: collection + width: 225 + horizontalAlignment: Text.AlignJustify + color: theme.textColor + placeholderText: qsTr("Collection name...") + placeholderTextColor: theme.mutedTextColor + ToolTip.text: qsTr("Name of the collection to add (Required)") + ToolTip.visible: hovered + Accessible.role: Accessible.EditableText + Accessible.name: collection.text + Accessible.description: ToolTip.text + function showError() { + collection.placeholderTextColor = theme.textErrorColor } - if (root.folder_path === "" || !folderEdit.isValid) { - isError = true; - folderEdit.showError(); + onTextChanged: { + collection.placeholderTextColor = theme.mutedTextColor + } + } + + MyDirectoryField { + id: folderEdit + Layout.fillWidth: true + text: root.folder_path + placeholderText: qsTr("Folder path...") + placeholderTextColor: theme.mutedTextColor + ToolTip.text: qsTr("Folder path to documents (Required)") + ToolTip.visible: hovered + function showError() { + folderEdit.placeholderTextColor = theme.textErrorColor + } + onTextChanged: { + folderEdit.placeholderTextColor = theme.mutedTextColor + } + } + + MyButton { + id: browseButton + text: qsTr("Browse") + onClicked: { + folderDialog.open(); + } + } + + MyButton { + id: addButton + text: qsTr("Add") + Accessible.role: Accessible.Button + Accessible.name: text + Accessible.description: qsTr("Add button") + onClicked: { + var isError = false; + if (root.collection === "") { + isError = true; + collection.showError(); + } + if (root.folder_path === "" || !folderEdit.isValid) { + isError = true; + folderEdit.showError(); + } + if (isError) + return; + LocalDocs.addFolder(root.collection, root.folder_path) + root.collection = "" + root.folder_path = "" + collection.clear() } - if (isError) - return; - LocalDocs.addFolder(root.collection, root.folder_path) - root.collection = "" - root.folder_path = "" - collection.clear() } } } - ScrollView { - id: scrollView - Layout.fillWidth: true - Layout.bottomMargin: 10 - clip: true - contentHeight: 300 - ScrollBar.vertical.policy: ScrollBar.AlwaysOn - - background: Rectangle { - color: theme.backgroundLighter - } - - ListView { - id: listView + ColumnLayout { + spacing: 0 + Repeater { model: LocalDocs.localDocsModel - boundsBehavior: Flickable.StopAtBounds delegate: Rectangle { id: item - width: listView.width + Layout.fillWidth: true height: buttons.height + 20 - color: index % 2 === 0 ? theme.backgroundLight : theme.backgroundLighter + color: index % 2 === 0 ? theme.backgroundDark : theme.backgroundDarker property bool removing: false Text { @@ -136,6 +135,7 @@ MySettingsTab { Text { id: folderId anchors.left: collectionId.right + anchors.right: buttons.left anchors.margins: 20 anchors.verticalCenter: parent.verticalCenter text: folder_path @@ -170,124 +170,99 @@ MySettingsTab { } } - GridLayout { - id: gridLayout + Rectangle { Layout.fillWidth: true - columns: 3 - rowSpacing: 10 - columnSpacing: 10 + height: 1 + color: theme.tabBorder + } + } + advancedSettings: GridLayout { + id: gridLayout + columns: 3 + rowSpacing: 10 + columnSpacing: 10 - Rectangle { - Layout.row: 0 - Layout.column: 0 - Layout.fillWidth: true - Layout.columnSpan: 3 - height: 1 - color: theme.dialogBorder + Rectangle { + Layout.row: 3 + Layout.column: 0 + Layout.fillWidth: true + Layout.columnSpan: 3 + height: 1 + color: theme.tabBorder + } + + Label { + id: chunkLabel + Layout.row: 1 + Layout.column: 0 + color: theme.textColor + text: qsTr("Document snippet size (characters):") + } + + MyTextField { + id: chunkSizeTextField + Layout.row: 1 + Layout.column: 1 + ToolTip.text: qsTr("Number of characters per document snippet.\nNOTE: larger numbers increase likelihood of factual responses, but also result in slower generation.") + ToolTip.visible: hovered + text: MySettings.localDocsChunkSize + validator: IntValidator { + bottom: 1 } - - Rectangle { - Layout.row: 3 - Layout.column: 0 - Layout.fillWidth: true - Layout.columnSpan: 3 - height: 1 - color: theme.dialogBorder - } - - // This is here just to stretch out the third column - Rectangle { - Layout.row: 3 - Layout.column: 2 - Layout.fillWidth: true - height: 1 - color: theme.dialogBorder - } - - Label { - id: chunkLabel - Layout.row: 1 - Layout.column: 0 - color: theme.textColor - text: qsTr("Document snippet size (characters):") - } - - MyTextField { - id: chunkSizeTextField - Layout.row: 1 - Layout.column: 1 - ToolTip.text: qsTr("Number of characters per document snippet.\nNOTE: larger numbers increase likelihood of factual responses, but also result in slower generation.") - ToolTip.visible: hovered - text: MySettings.localDocsChunkSize - validator: IntValidator { - bottom: 1 - } - onEditingFinished: { - var val = parseInt(text) - if (!isNaN(val)) { - MySettings.localDocsChunkSize = val - focus = false - } else { - text = MySettings.localDocsChunkSize - } + onEditingFinished: { + var val = parseInt(text) + if (!isNaN(val)) { + MySettings.localDocsChunkSize = val + focus = false + } else { + text = MySettings.localDocsChunkSize } } + } - Label { - id: contextItemsPerPrompt - Layout.row: 2 - Layout.column: 0 - color: theme.textColor - text: qsTr("Document snippets per prompt:") + Label { + id: contextItemsPerPrompt + Layout.row: 2 + Layout.column: 0 + color: theme.textColor + text: qsTr("Document snippets per prompt:") + } + + MyTextField { + Layout.row: 2 + Layout.column: 1 + ToolTip.text: qsTr("Best N matches of retrieved document snippets to add to the context for prompt.\nNOTE: larger numbers increase likelihood of factual responses, but also result in slower generation.") + ToolTip.visible: hovered + text: MySettings.localDocsRetrievalSize + validator: IntValidator { + bottom: 1 } - - MyTextField { - Layout.row: 2 - Layout.column: 1 - ToolTip.text: qsTr("Best N matches of retrieved document snippets to add to the context for prompt.\nNOTE: larger numbers increase likelihood of factual responses, but also result in slower generation.") - ToolTip.visible: hovered - text: MySettings.localDocsRetrievalSize - validator: IntValidator { - bottom: 1 - } - onEditingFinished: { - var val = parseInt(text) - if (!isNaN(val)) { - MySettings.localDocsRetrievalSize = val - focus = false - } else { - text = MySettings.localDocsRetrievalSize - } + onEditingFinished: { + var val = parseInt(text) + if (!isNaN(val)) { + MySettings.localDocsRetrievalSize = val + focus = false + } else { + text = MySettings.localDocsRetrievalSize } } + } + Item { + Layout.row: 1 + Layout.column: 2 + Layout.rowSpan: 2 + Layout.fillWidth: true + Layout.alignment: Qt.AlignTop + Layout.minimumHeight: warningLabel.height Label { id: warningLabel - Layout.row: 1 - Layout.column: 2 - Layout.rowSpan: 2 - Layout.maximumWidth: 520 - Layout.alignment: Qt.AlignTop + width: parent.width color: theme.textErrorColor wrapMode: Text.WordWrap text: qsTr("Warning: Advanced usage only. Values too large may cause localdocs failure, extremely slow responses or failure to respond at all. Roughly speaking, the {N chars x N snippets} are added to the model's context window. More info here.") onLinkActivated: function(link) { Qt.openUrlExternally(link) } } - - MyButton { - id: restoreDefaultsButton - Layout.row: 4 - Layout.column: 1 - Layout.columnSpan: 2 - Layout.fillWidth: true - text: qsTr("Restore Defaults") - Accessible.role: Accessible.Button - Accessible.name: text - Accessible.description: qsTr("Restores the settings dialog to a default state") - onClicked: { - MySettings.restoreLocalDocsDefaults(); - } - } } } } diff --git a/gpt4all-chat/qml/ModelDownloaderDialog.qml b/gpt4all-chat/qml/ModelDownloaderDialog.qml index a97c17d1..e2839c76 100644 --- a/gpt4all-chat/qml/ModelDownloaderDialog.qml +++ b/gpt4all-chat/qml/ModelDownloaderDialog.qml @@ -83,7 +83,7 @@ Dialog { Text { textFormat: Text.StyledText - text: "

    " + (name !== "" ? name : filename) + "

    " + text: "

    " + name + "

    " Layout.row: 0 Layout.column: 0 Layout.topMargin: 20 @@ -329,12 +329,14 @@ Dialog { Layout.topMargin: 20 Layout.leftMargin: 20 Layout.minimumWidth: 150 + Layout.maximumWidth: textMetrics.width + 25 Layout.alignment: Qt.AlignTop | Qt.AlignHCenter color: theme.textColor background: Rectangle { color: theme.backgroundLighter radius: 10 } + wrapMode: Text.WrapAnywhere function showError() { openaiKey.placeholderTextColor = theme.textErrorColor } @@ -346,6 +348,11 @@ Dialog { Accessible.role: Accessible.EditableText Accessible.name: placeholderText Accessible.description: qsTr("Whether the file hash is being calculated") + TextMetrics { + id: textMetrics + font: openaiKey.font + text: openaiKey.placeholderText + } } } } diff --git a/gpt4all-chat/qml/ModelSettings.qml b/gpt4all-chat/qml/ModelSettings.qml index 1db9a42e..7007c010 100644 --- a/gpt4all-chat/qml/ModelSettings.qml +++ b/gpt4all-chat/qml/ModelSettings.qml @@ -3,27 +3,665 @@ import QtQuick import QtQuick.Controls import QtQuick.Controls.Basic import QtQuick.Layouts -import localdocs +import modellist import mysettings MySettingsTab { - title: qsTr("Models") + onRestoreDefaultsClicked: { + MySettings.restoreModelDefaults(root.currentModelInfo); + } + title: qsTr("Model/Character Settings") contentItem: GridLayout { - id: generationSettingsTabInner - columns: 2 + id: root + columns: 3 rowSpacing: 10 columnSpacing: 10 - MyButton { - Layout.row: 8 - Layout.column: 1 + property var currentModelName: comboBox.currentText + property var currentModelId: comboBox.currentValue + property var currentModelInfo: ModelList.modelInfo(root.currentModelId) + + Label { + id: label + Layout.row: 0 + Layout.column: 0 + text: qsTr("Model/Character:") + color: theme.textColor + } + + RowLayout { Layout.fillWidth: true - text: qsTr("Restore Defaults") - Accessible.description: qsTr("Restores the settings dialog to a default state") - onClicked: { - MySettings.restoreGenerationDefaults(); - templateTextArea.text = MySettings.promptTemplate + Layout.row: 1 + Layout.column: 0 + Layout.columnSpan: 2 + height: label.height + 20 + spacing: 10 + + MyComboBox { + id: comboBox + Layout.fillWidth: true + model: ModelList.installedModels + valueRole: "id" + textRole: "name" + currentIndex: 0 + contentItem: Text { + leftPadding: 10 + rightPadding: 20 + text: comboBox.currentText + font: comboBox.font + color: theme.textColor + verticalAlignment: Text.AlignVCenter + elide: Text.ElideRight + } + delegate: ItemDelegate { + width: comboBox.width + contentItem: Text { + text: name + color: theme.textColor + font: comboBox.font + elide: Text.ElideRight + verticalAlignment: Text.AlignVCenter + } + background: Rectangle { + color: highlighted ? theme.backgroundLight : theme.backgroundDark + } + highlighted: comboBox.highlightedIndex === index + } } + + MyButton { + id: cloneButton + text: qsTr("Clone") + onClicked: { + var id = ModelList.clone(root.currentModelInfo); + comboBox.currentIndex = comboBox.indexOfValue(id); + } + } + + MyButton { + id: removeButton + enabled: root.currentModelInfo.isClone + text: qsTr("Remove") + onClicked: { + ModelList.remove(root.currentModelInfo); + comboBox.currentIndex = 0; + } + } + } + + RowLayout { + Layout.row: 2 + Layout.column: 0 + Layout.topMargin: 15 + spacing: 10 + Label { + id: uniqueNameLabel + text: qsTr("Unique Name:") + color: theme.textColor + } + Label { + id: uniqueNameLabelHelp + visible: false + text: qsTr("Must contain a non-empty unique name that does not match any existing model/character.") + color: theme.textErrorColor + wrapMode: TextArea.Wrap + } + } + + MyTextField { + id: uniqueNameField + text: root.currentModelName + enabled: root.currentModelInfo.isClone || root.currentModelInfo.description === "" + color: enabled ? theme.textColor : theme.mutedTextColor + Layout.row: 3 + Layout.column: 0 + Layout.columnSpan: 2 + Layout.fillWidth: true + Connections { + target: MySettings + function onNameChanged() { + uniqueNameField.text = root.currentModelInfo.name; + } + } + Connections { + target: root + function onCurrentModelInfoChanged() { + uniqueNameField.text = root.currentModelInfo.name; + } + } + onTextChanged: { + if (text !== "" && ModelList.isUniqueName(text)) { + MySettings.setModelName(root.currentModelInfo, text); + } + uniqueNameLabelHelp.visible = root.currentModelInfo.name !== "" && + (text === "" || (text !== root.currentModelInfo.name && !ModelList.isUniqueName(text))); + } + } + + Label { + text: qsTr("Model File:") + color: theme.textColor + Layout.row: 4 + Layout.column: 0 + Layout.topMargin: 15 + } + + MyTextField { + text: root.currentModelInfo.filename + enabled: false + color: enabled ? theme.textColor : theme.mutedTextColor + Layout.row: 5 + Layout.column: 0 + Layout.columnSpan: 2 + Layout.fillWidth: true + } + + Label { + visible: !root.currentModelInfo.isChatGPT + text: qsTr("System Prompt:") + color: theme.textColor + Layout.row: 6 + Layout.column: 0 + Layout.topMargin: 15 + } + + Rectangle { + id: systemPrompt + visible: !root.currentModelInfo.isChatGPT + Layout.row: 7 + Layout.column: 0 + Layout.columnSpan: 2 + Layout.fillWidth: true + color: "transparent" + Layout.minimumHeight: Math.max(150, systemPromptArea.contentHeight + 20) + TextArea { + id: systemPromptArea + anchors.fill: parent + text: root.currentModelInfo.systemPrompt + color: theme.textColor + background: Rectangle { + implicitWidth: 150 + color: theme.backgroundDark + radius: 10 + } + padding: 10 + wrapMode: TextArea.Wrap + Connections { + target: MySettings + function onSystemPromptChanged() { + systemPromptArea.text = root.currentModelInfo.systemPrompt; + } + } + Connections { + target: root + function onCurrentModelInfoChanged() { + systemPromptArea.text = root.currentModelInfo.systemPrompt; + } + } + onTextChanged: { + MySettings.setModelSystemPrompt(root.currentModelInfo, text) + } + bottomPadding: 10 + Accessible.role: Accessible.EditableText + ToolTip.text: qsTr("The systemPrompt allows instructions to the model at the beginning of a chat.\nNOTE: A longer, detailed system prompt can lead to higher quality answers, but can also slow down generation.") + ToolTip.visible: hovered + } + } + + RowLayout { + Layout.row: 8 + Layout.column: 0 + Layout.columnSpan: 2 + Layout.topMargin: 15 + spacing: 10 + Label { + id: promptTemplateLabel + text: qsTr("Prompt Template:") + color: theme.textColor + } + Label { + id: promptTemplateLabelHelp + text: qsTr("Must contain the string \"%1\" to be replaced with the user's input.") + color: theme.textErrorColor + visible: templateTextArea.text.indexOf("%1") === -1 + wrapMode: TextArea.Wrap + } + } + + Rectangle { + id: promptTemplate + Layout.row: 9 + Layout.column: 0 + Layout.columnSpan: 2 + Layout.fillWidth: true + Layout.minimumHeight: Math.max(150, templateTextArea.contentHeight + 20) + color: "transparent" + clip: true + TextArea { + id: templateTextArea + anchors.fill: parent + text: root.currentModelInfo.promptTemplate + color: theme.textColor + background: Rectangle { + implicitWidth: 150 + color: theme.backgroundDark + radius: 10 + } + padding: 10 + wrapMode: TextArea.Wrap + Connections { + target: MySettings + function onPromptTemplateChanged() { + templateTextArea.text = root.currentModelInfo.promptTemplate; + } + } + Connections { + target: root + function onCurrentModelInfoChanged() { + templateTextArea.text = root.currentModelInfo.promptTemplate; + } + } + onTextChanged: { + if (templateTextArea.text.indexOf("%1") !== -1) { + MySettings.setModelPromptTemplate(root.currentModelInfo, text) + } + } + bottomPadding: 10 + Accessible.role: Accessible.EditableText + Accessible.name: promptTemplateLabel.text + Accessible.description: promptTemplateLabelHelp.text + ToolTip.text: qsTr("The prompt template partially determines how models will respond to prompts.\nNOTE: A longer, detailed template can lead to higher quality answers, but can also slow down generation.") + ToolTip.visible: hovered + } + } + + Rectangle { + id: optionalImageRect + visible: false // FIXME: for later + Layout.row: 2 + Layout.column: 1 + Layout.rowSpan: 5 + Layout.alignment: Qt.AlignHCenter + Layout.fillHeight: true + Layout.maximumWidth: height + Layout.topMargin: 35 + Layout.bottomMargin: 35 + Layout.leftMargin: 35 + width: 3000 + border.width: 1 + border.color: theme.tabBorder + radius: 10 + color: "transparent" + Item { + anchors.centerIn: parent + height: childrenRect.height + Image { + id: img + anchors.horizontalCenter: parent.horizontalCenter + width: 100 + height: 100 + source: "qrc:/gpt4all/icons/image.svg" + } + Text { + text: qsTr("Add\noptional image") + anchors.top: img.bottom + anchors.horizontalCenter: parent.horizontalCenter + wrapMode: TextArea.Wrap + horizontalAlignment: Qt.AlignHCenter + color: theme.mutedTextColor + } + } + } + + Label { + text: qsTr("Generation Settings") + color: theme.textColor + Layout.row: 10 + Layout.column: 0 + Layout.columnSpan: 2 + Layout.topMargin: 15 + Layout.alignment: Qt.AlignHCenter + Layout.minimumWidth: promptTemplate.width + horizontalAlignment: Qt.AlignHCenter + font.pixelSize: theme.fontSizeLarger + font.bold: true + } + + GridLayout { + Layout.row: 11 + Layout.column: 0 + Layout.columnSpan: 2 + Layout.topMargin: 15 + Layout.fillWidth: true + Layout.minimumWidth: promptTemplate.width + columns: 4 + rowSpacing: 10 + columnSpacing: 10 + + Label { + id: tempLabel + text: qsTr("Temperature:") + color: theme.textColor + Layout.row: 0 + Layout.column: 0 + } + + MyTextField { + id: temperatureField + text: root.currentModelInfo.temperature + color: theme.textColor + ToolTip.text: qsTr("Temperature increases the chances of choosing less likely tokens.\nNOTE: Higher temperature gives more creative but less predictable outputs.") + ToolTip.visible: hovered + Layout.row: 0 + Layout.column: 1 + validator: DoubleValidator { + locale: "C" + } + Connections { + target: MySettings + function onTemperatureChanged() { + temperatureField.text = root.currentModelInfo.temperature; + } + } + Connections { + target: root + function onCurrentModelInfoChanged() { + temperatureField.text = root.currentModelInfo.temperature; + } + } + onEditingFinished: { + var val = parseFloat(text) + if (!isNaN(val)) { + MySettings.setModelTemperature(root.currentModelInfo, val) + focus = false + } else { + text = root.currentModelInfo.temperature + } + } + Accessible.role: Accessible.EditableText + Accessible.name: tempLabel.text + Accessible.description: ToolTip.text + } + Label { + id: topPLabel + text: qsTr("Top P:") + color: theme.textColor + Layout.row: 0 + Layout.column: 2 + } + MyTextField { + id: topPField + text: root.currentModelInfo.topP + color: theme.textColor + ToolTip.text: qsTr("Only the most likely tokens up to a total probability of top_p can be chosen.\nNOTE: Prevents choosing highly unlikely tokens, aka Nucleus Sampling") + ToolTip.visible: hovered + Layout.row: 0 + Layout.column: 3 + validator: DoubleValidator { + locale: "C" + } + Connections { + target: MySettings + function onTopPChanged() { + topPField.text = root.currentModelInfo.topP; + } + } + Connections { + target: root + function onCurrentModelInfoChanged() { + topPField.text = root.currentModelInfo.topP; + } + } + onEditingFinished: { + var val = parseFloat(text) + if (!isNaN(val)) { + MySettings.setModelTopP(root.currentModelInfo, val) + focus = false + } else { + text = root.currentModelInfo.topP + } + } + Accessible.role: Accessible.EditableText + Accessible.name: topPLabel.text + Accessible.description: ToolTip.text + } + Label { + id: topKLabel + visible: !root.currentModelInfo.isChatGPT + text: qsTr("Top K:") + color: theme.textColor + Layout.row: 1 + Layout.column: 0 + } + MyTextField { + id: topKField + visible: !root.currentModelInfo.isChatGPT + text: root.currentModelInfo.topK + color: theme.textColor + ToolTip.text: qsTr("Only the top K most likely tokens will be chosen from") + ToolTip.visible: hovered + Layout.row: 1 + Layout.column: 1 + validator: IntValidator { + bottom: 1 + } + Connections { + target: MySettings + function onTopKChanged() { + topKField.text = root.currentModelInfo.topK; + } + } + Connections { + target: root + function onCurrentModelInfoChanged() { + topKField.text = root.currentModelInfo.topK; + } + } + onEditingFinished: { + var val = parseInt(text) + if (!isNaN(val)) { + MySettings.setModelTopK(root.currentModelInfo, val) + focus = false + } else { + text = root.currentModelInfo.topK + } + } + Accessible.role: Accessible.EditableText + Accessible.name: topKLabel.text + Accessible.description: ToolTip.text + } + Label { + id: maxLengthLabel + visible: !root.currentModelInfo.isChatGPT + text: qsTr("Max Length:") + color: theme.textColor + Layout.row: 1 + Layout.column: 2 + } + MyTextField { + id: maxLengthField + visible: !root.currentModelInfo.isChatGPT + text: root.currentModelInfo.maxLength + color: theme.textColor + ToolTip.text: qsTr("Maximum length of response in tokens") + ToolTip.visible: hovered + Layout.row: 1 + Layout.column: 3 + validator: IntValidator { + bottom: 1 + } + Connections { + target: MySettings + function onMaxLengthChanged() { + maxLengthField.text = root.currentModelInfo.maxLength; + } + } + Connections { + target: root + function onCurrentModelInfoChanged() { + maxLengthField.text = root.currentModelInfo.maxLength; + } + } + onEditingFinished: { + var val = parseInt(text) + if (!isNaN(val)) { + MySettings.setModelMaxLength(root.currentModelInfo, val) + focus = false + } else { + text = root.currentModelInfo.maxLength + } + } + Accessible.role: Accessible.EditableText + Accessible.name: maxLengthLabel.text + Accessible.description: ToolTip.text + } + + Label { + id: batchSizeLabel + visible: !root.currentModelInfo.isChatGPT + text: qsTr("Prompt Batch Size:") + color: theme.textColor + Layout.row: 2 + Layout.column: 0 + } + MyTextField { + id: batchSizeField + visible: !root.currentModelInfo.isChatGPT + text: root.currentModelInfo.promptBatchSize + color: theme.textColor + ToolTip.text: qsTr("Amount of prompt tokens to process at once.\nNOTE: Higher values can speed up reading prompts but will use more RAM") + ToolTip.visible: hovered + Layout.row: 2 + Layout.column: 1 + validator: IntValidator { + bottom: 1 + } + Connections { + target: MySettings + function onPromptBatchSizeChanged() { + batchSizeField.text = root.currentModelInfo.promptBatchSize; + } + } + Connections { + target: root + function onCurrentModelInfoChanged() { + batchSizeField.text = root.currentModelInfo.promptBatchSize; + } + } + onEditingFinished: { + var val = parseInt(text) + if (!isNaN(val)) { + MySettings.setModelPromptBatchSize(root.currentModelInfo, val) + focus = false + } else { + text = root.currentModelInfo.promptBatchSize + } + } + Accessible.role: Accessible.EditableText + Accessible.name: batchSizeLabel.text + Accessible.description: ToolTip.text + } + Label { + id: repeatPenaltyLabel + visible: !root.currentModelInfo.isChatGPT + text: qsTr("Repeat Penalty:") + color: theme.textColor + Layout.row: 2 + Layout.column: 2 + } + MyTextField { + id: repeatPenaltyField + visible: !root.currentModelInfo.isChatGPT + text: root.currentModelInfo.repeatPenalty + color: theme.textColor + ToolTip.text: qsTr("Amount to penalize repetitiveness of the output") + ToolTip.visible: hovered + Layout.row: 2 + Layout.column: 3 + validator: DoubleValidator { + locale: "C" + } + Connections { + target: MySettings + function onRepeatPenaltyChanged() { + repeatPenaltyField.text = root.currentModelInfo.repeatPenalty; + } + } + Connections { + target: root + function onCurrentModelInfoChanged() { + repeatPenaltyField.text = root.currentModelInfo.repeatPenalty; + } + } + onEditingFinished: { + var val = parseFloat(text) + if (!isNaN(val)) { + MySettings.setModelRepeatPenalty(root.currentModelInfo, val) + focus = false + } else { + text = root.currentModelInfo.repeatPenalty + } + } + Accessible.role: Accessible.EditableText + Accessible.name: repeatPenaltyLabel.text + Accessible.description: ToolTip.text + } + Label { + id: repeatPenaltyTokensLabel + visible: !root.currentModelInfo.isChatGPT + text: qsTr("Repeat Penalty Tokens:") + color: theme.textColor + Layout.row: 3 + Layout.column: 0 + } + MyTextField { + id: repeatPenaltyTokenField + visible: !root.currentModelInfo.isChatGPT + text: root.currentModelInfo.repeatPenaltyTokens + color: theme.textColor + ToolTip.text: qsTr("How far back in output to apply repeat penalty") + ToolTip.visible: hovered + Layout.row: 3 + Layout.column: 1 + validator: IntValidator { + bottom: 1 + } + Connections { + target: MySettings + function onRepeatPenaltyTokensChanged() { + repeatPenaltyTokenField.text = root.currentModelInfo.repeatPenaltyTokens; + } + } + Connections { + target: root + function onCurrentModelInfoChanged() { + repeatPenaltyTokenField.text = root.currentModelInfo.repeatPenaltyTokens; + } + } + onEditingFinished: { + var val = parseInt(text) + if (!isNaN(val)) { + MySettings.setModelRepeatPenaltyTokens(root.currentModelInfo, val) + focus = false + } else { + text = root.currentModelInfo.repeatPenaltyTokens + } + } + Accessible.role: Accessible.EditableText + Accessible.name: repeatPenaltyTokensLabel.text + Accessible.description: ToolTip.text + } + } + + Rectangle { + Layout.row: 12 + Layout.column: 0 + Layout.columnSpan: 2 + Layout.topMargin: 15 + Layout.fillWidth: true + Layout.minimumWidth: promptTemplate.width + height: 1 + color: theme.tabBorder } } } \ No newline at end of file diff --git a/gpt4all-chat/qml/MyButton.qml b/gpt4all-chat/qml/MyButton.qml index 8710008a..44d70770 100644 --- a/gpt4all-chat/qml/MyButton.qml +++ b/gpt4all-chat/qml/MyButton.qml @@ -17,7 +17,7 @@ Button { border.color: myButton.down ? theme.backgroundLightest : theme.buttonBorder border.width: 2 radius: 10 - color: myButton.hovered ? theme.backgroundLighter : theme.backgroundLight + color: myButton.hovered ? theme.backgroundDark : theme.backgroundDarkest } Accessible.role: Accessible.Button Accessible.name: text diff --git a/gpt4all-chat/qml/MyComboBox.qml b/gpt4all-chat/qml/MyComboBox.qml index 736c5fb4..e55a5b04 100644 --- a/gpt4all-chat/qml/MyComboBox.qml +++ b/gpt4all-chat/qml/MyComboBox.qml @@ -9,14 +9,13 @@ ComboBox { padding: 10 Accessible.role: Accessible.ComboBox contentItem: Text { - anchors.horizontalCenter: parent.horizontalCenter + id: text leftPadding: 10 rightPadding: 20 text: comboBox.displayText font: comboBox.font color: theme.textColor verticalAlignment: Text.AlignVCenter - horizontalAlignment: Text.AlignHCenter elide: Text.ElideRight } delegate: ItemDelegate { diff --git a/gpt4all-chat/qml/MyDirectoryField.qml b/gpt4all-chat/qml/MyDirectoryField.qml index e5dbe1b0..ed61ca56 100644 --- a/gpt4all-chat/qml/MyDirectoryField.qml +++ b/gpt4all-chat/qml/MyDirectoryField.qml @@ -11,7 +11,7 @@ TextField { color: text === "" || isValid ? theme.textColor : theme.textErrorColor background: Rectangle { implicitWidth: 150 - color: theme.backgroundLighter + color: theme.backgroundDark radius: 10 } } diff --git a/gpt4all-chat/qml/MySettingsStack.qml b/gpt4all-chat/qml/MySettingsStack.qml index 69321fd7..fba9fe07 100644 --- a/gpt4all-chat/qml/MySettingsStack.qml +++ b/gpt4all-chat/qml/MySettingsStack.qml @@ -12,71 +12,56 @@ Item { id: theme } + property alias title: titleLabel.text property ListModel tabTitlesModel: ListModel { } property list tabs: [ ] - Canvas { - id: canvas - anchors.fill: parent - contextType: "2d" - onPaint: { - var context = getContext("2d"); - context.reset(); - context.lineWidth = 2; - context.moveTo(stackLayout.x, stackLayout.y); - context.lineTo(stackLayout.x, stackLayout.y + stackLayout.height); - context.lineTo(stackLayout.x + stackLayout.width, stackLayout.y + stackLayout.height); - context.lineTo(stackLayout.x + stackLayout.width, stackLayout.y); - context.lineTo(/*settingsTabBar.currentItem.x + */settingsTabBar.currentItem.width, stackLayout.y); - context.strokeStyle = theme.tabBorder; - context.stroke(); - } + Label { + id: titleLabel + anchors.top: parent.top + anchors.horizontalCenter: parent.horizontalCenter + color: theme.textColor + padding: 10 + font.bold: true + font.pixelSize: theme.fontSizeLarger + } + + Rectangle { + anchors.top: titleLabel.bottom + anchors.leftMargin: 15 + anchors.rightMargin: 15 + anchors.left: parent.left + anchors.right: parent.right + height: 1 + color: theme.tabBorder } TabBar { id: settingsTabBar - width: parent.width / 1.25 + anchors.top: titleLabel.bottom + anchors.topMargin: 15 + anchors.horizontalCenter: parent.horizontalCenter + width: parent.width / 1.75 z: 200 visible: tabTitlesModel.count > 1 + background: Rectangle { + color: "transparent" + } Repeater { model: settingsStack.tabTitlesModel TabButton { id: tabButton + padding: 10 contentItem: IconLabel { color: theme.textColor font.bold: tabButton.checked text: model.title } background: Rectangle { - color: tabButton.checked ? theme.backgroundDarkest : theme.backgroundLight -// Rectangle { -// anchors.top: parent.top -// anchors.left: parent.left -// anchors.right: parent.right -// height: tabButton.checked -// color: theme.tabBorder -// } -// Rectangle { -// anchors.bottom: parent.bottom -// anchors.left: parent.left -// anchors.right: parent.right -// height: !tabButton.checked -// color: theme.tabBorder -// } -// Rectangle { -// anchors.top: parent.top -// anchors.bottom: parent.bottom -// anchors.left: parent.left -// width: tabButton.checked -// color: theme.tabBorder -// } -// Rectangle { -// anchors.top: parent.top -// anchors.bottom: parent.bottom -// anchors.right: parent.right -// width: tabButton.checked -// color: theme.tabBorder -// } + color: "transparent" + border.width: 1 + border.color: tabButton.checked ? theme.tabBorder : "transparent" + radius: 10 } Accessible.role: Accessible.Button Accessible.name: model.title @@ -84,9 +69,31 @@ Item { } } + Rectangle { + id: dividerTabBar + visible: tabTitlesModel.count > 1 + anchors.top: settingsTabBar.bottom + anchors.topMargin: 15 + anchors.bottomMargin: 15 + anchors.leftMargin: 15 + anchors.rightMargin: 15 + anchors.left: parent.left + anchors.right: parent.right + height: 1 + color: theme.tabBorder + } + + Rectangle { + anchors.fill: parent + color: "transparent" + radius: 10 + border.width: 1 + border.color: theme.tabBorder + } + StackLayout { id: stackLayout - anchors.top: tabTitlesModel.count > 1 ? settingsTabBar.bottom : parent.top + anchors.top: tabTitlesModel.count > 1 ? dividerTabBar.bottom : titleLabel.bottom anchors.left: parent.left anchors.right: parent.right anchors.bottom: parent.bottom diff --git a/gpt4all-chat/qml/MySettingsTab.qml b/gpt4all-chat/qml/MySettingsTab.qml index 6c35ad91..9593564e 100644 --- a/gpt4all-chat/qml/MySettingsTab.qml +++ b/gpt4all-chat/qml/MySettingsTab.qml @@ -5,42 +5,87 @@ import QtQuick.Controls.Basic import QtQuick.Layouts Item { + id: root property string title: "" property Item contentItem: null + property Item advancedSettings: null + signal restoreDefaultsClicked onContentItemChanged: function() { if (contentItem) { - contentItem.parent = tabInner; - contentItem.anchors.left = tabInner.left; - contentItem.anchors.right = tabInner.right; + contentItem.parent = contentInner; + contentItem.anchors.left = contentInner.left; + contentItem.anchors.right = contentInner.right; + } + } + + onAdvancedSettingsChanged: function() { + if (advancedSettings) { + advancedSettings.parent = advancedInner; + advancedSettings.anchors.left = advancedInner.left; + advancedSettings.anchors.right = advancedInner.right; } } ScrollView { - id: root width: parent.width height: parent.height padding: 15 rightPadding: 20 contentWidth: availableWidth - contentHeight: tabInner.height + contentHeight: innerColumn.height ScrollBar.vertical.policy: ScrollBar.AlwaysOn Theme { id: theme } -// background: Rectangle { -// color: 'transparent' -// border.color: theme.tabBorder -// border.width: 1 -// radius: 10 -// } - - Column { - id: tabInner + ColumnLayout { + id: innerColumn anchors.left: parent.left anchors.right: parent.right + anchors.margins: 15 + spacing: 10 + Column { + id: contentInner + Layout.fillWidth: true + } + + Column { + id: advancedInner + visible: false + Layout.fillWidth: true + } + + Item { + Layout.fillWidth: true + height: restoreDefaultsButton.height + MyButton { + id: restoreDefaultsButton + anchors.left: parent.left + width: implicitWidth + text: qsTr("Restore Defaults") + Accessible.role: Accessible.Button + Accessible.name: text + Accessible.description: qsTr("Restores the settings dialog to a default state") + onClicked: { + root.restoreDefaultsClicked(); + } + } + MyButton { + id: advancedSettingsButton + anchors.right: parent.right + visible: root.advancedSettings + width: implicitWidth + text: !advancedInner.visible ? qsTr("Advanced Settings") : qsTr("Hide Advanced Settings") + Accessible.role: Accessible.Button + Accessible.name: text + Accessible.description: qsTr("Shows/hides the advanced settings") + onClicked: { + advancedInner.visible = !advancedInner.visible; + } + } + } } } } diff --git a/gpt4all-chat/qml/MyTextField.qml b/gpt4all-chat/qml/MyTextField.qml index 6bef8936..730d8021 100644 --- a/gpt4all-chat/qml/MyTextField.qml +++ b/gpt4all-chat/qml/MyTextField.qml @@ -8,7 +8,7 @@ TextField { padding: 10 background: Rectangle { implicitWidth: 150 - color: theme.backgroundLighter + color: theme.backgroundDark radius: 10 } } \ No newline at end of file diff --git a/gpt4all-chat/qml/SettingsDialog.qml b/gpt4all-chat/qml/SettingsDialog.qml index 92991e44..491985e8 100644 --- a/gpt4all-chat/qml/SettingsDialog.qml +++ b/gpt4all-chat/qml/SettingsDialog.qml @@ -14,7 +14,6 @@ import mysettings Dialog { id: settingsDialog modal: true - opacity: 0.9 padding: 20 bottomPadding: 30 background: Rectangle { @@ -38,7 +37,7 @@ Dialog { ListModel { id: stacksModel ListElement { - title: "Generation" + title: "Models" } ListElement { title: "Application" @@ -67,7 +66,6 @@ Dialog { id: item width: listView.width height: titleLabel.height + 25 -// color: index % 2 === 0 ? theme.backgroundLight : theme.backgroundLighter color: "transparent" border.color: theme.backgroundLighter border.width: index == listView.currentIndex ? 1 : 0 @@ -78,6 +76,7 @@ Dialog { anchors.verticalCenter: parent.verticalCenter anchors.left: parent.left anchors.margins: 20 + font.bold: index == listView.currentIndex text: title elide: Text.ElideRight color: theme.textColor @@ -101,19 +100,21 @@ Dialog { currentIndex: listView.currentIndex MySettingsStack { + title: qsTr("Model/Character Settings") tabs: [ - Component { ModelSettings { } }, - Component { GenerationSettings { } } + Component { ModelSettings { } } ] } MySettingsStack { + title: qsTr("Application General Settings") tabs: [ Component { ApplicationSettings { } } ] } MySettingsStack { + title: qsTr("LocalDocs Plugin (BETA) Settings") tabs: [ Component { LocalDocsSettings { } } ] diff --git a/gpt4all-chat/qml/Theme.qml b/gpt4all-chat/qml/Theme.qml index 79923762..90d4fc94 100644 --- a/gpt4all-chat/qml/Theme.qml +++ b/gpt4all-chat/qml/Theme.qml @@ -7,9 +7,9 @@ QtObject { property color textAccent: "#8e8ea0" property color mutedTextColor: backgroundLightest property color textErrorColor: "red" - property color backgroundDarkest: "#202123" - property color backgroundDarker: "#222326" - property color backgroundDark: "#242528" + property color backgroundDarkest: "#1c1f21" + property color backgroundDarker: "#1e2123" + property color backgroundDark: "#222527" property color backgroundLight: "#343541" property color backgroundLighter: "#444654" property color backgroundLightest: "#7d7d8e" diff --git a/gpt4all-chat/responsetext.cpp b/gpt4all-chat/responsetext.cpp index 1a4520fb..8dfa6562 100644 --- a/gpt4all-chat/responsetext.cpp +++ b/gpt4all-chat/responsetext.cpp @@ -120,6 +120,74 @@ static QVector pythonHighlightingRules() return highlightingRules; } +static QVector csharpHighlightingRules() +{ + static QVector highlightingRules; + if (highlightingRules.isEmpty()) { + + HighlightingRule rule; + + // Function call highlighting + QTextCharFormat functionCallFormat; + functionCallFormat.setForeground(functionCallColor); + rule.pattern = QRegularExpression("\\b(\\w+)\\s*(?=\\()"); + rule.format = functionCallFormat; + highlightingRules.append(rule); + + // Function definition highlighting + QTextCharFormat functionFormat; + functionFormat.setForeground(functionColor); + rule.pattern = QRegularExpression("\\bvoid|int|double|string|bool\\s+(\\w+)\\s*(?=\\()"); + rule.format = functionFormat; + highlightingRules.append(rule); + + // Number highlighting + QTextCharFormat numberFormat; + numberFormat.setForeground(numberColor); + rule.pattern = QRegularExpression("\\b[0-9]*\\.?[0-9]+\\b"); + rule.format = numberFormat; + highlightingRules.append(rule); + + // Keyword highlighting + QTextCharFormat keywordFormat; + keywordFormat.setForeground(keywordColor); + QStringList keywordPatterns = { + "\\bvoid\\b", "\\bint\\b", "\\bdouble\\b", "\\bstring\\b", "\\bbool\\b", + "\\bclass\\b", "\\bif\\b", "\\belse\\b", "\\bwhile\\b", "\\bfor\\b", + "\\breturn\\b", "\\bnew\\b", "\\bthis\\b", "\\bpublic\\b", "\\bprivate\\b", + "\\bprotected\\b", "\\bstatic\\b", "\\btrue\\b", "\\bfalse\\b", "\\bnull\\b", + "\\bnamespace\\b", "\\busing\\b", "\\btry\\b", "\\bcatch\\b", "\\bfinally\\b", + "\\bthrow\\b", "\\bvar\\b" + }; + + for (const QString &pattern : keywordPatterns) { + rule.pattern = QRegularExpression(pattern); + rule.format = keywordFormat; + highlightingRules.append(rule); + } + + // String highlighting + QTextCharFormat stringFormat; + stringFormat.setForeground(stringColor); + rule.pattern = QRegularExpression("\".*?\""); + rule.format = stringFormat; + highlightingRules.append(rule); + + // Single-line comment highlighting + QTextCharFormat commentFormat; + commentFormat.setForeground(commentColor); + rule.pattern = QRegularExpression("//[^\n]*"); + rule.format = commentFormat; + highlightingRules.append(rule); + + // Multi-line comment highlighting + rule.pattern = QRegularExpression("/\\*.*?\\*/"); + rule.format = commentFormat; + highlightingRules.append(rule); + } + return highlightingRules; +} + static QVector cppHighlightingRules() { static QVector highlightingRules; diff --git a/gpt4all-chat/server.cpp b/gpt4all-chat/server.cpp index 2b3a258f..09d8256a 100644 --- a/gpt4all-chat/server.cpp +++ b/gpt4all-chat/server.cpp @@ -14,11 +14,11 @@ static inline QJsonObject modelToJson(const ModelInfo &info) { QJsonObject model; - model.insert("id", info.name); + model.insert("id", info.name()); model.insert("object", "model"); model.insert("created", "who can keep track?"); model.insert("owned_by", "humanity"); - model.insert("root", info.name); + model.insert("root", info.name()); model.insert("parent", QJsonValue::Null); QJsonArray permissions; @@ -106,7 +106,7 @@ void Server::start() if (!info.installed) continue; - if (model == info.name) { + if (model == info.name()) { object = modelToJson(info); break; } @@ -173,7 +173,7 @@ QHttpServerResponse Server::handleCompletionRequest(const QHttpServerRequest &re for (const ModelInfo &info : modelList) { if (!info.installed) continue; - if (modelRequested == info.name || modelRequested == info.filename) { + if (modelRequested == info.name() || modelRequested == info.filename()) { modelInfo = info; break; } @@ -284,32 +284,28 @@ QHttpServerResponse Server::handleCompletionRequest(const QHttpServerRequest &re // load the new model if necessary setShouldBeLoaded(true); - if (modelInfo.filename.isEmpty()) { + if (modelInfo.filename().isEmpty()) { std::cerr << "ERROR: couldn't load default model " << modelRequested.toStdString() << std::endl; return QHttpServerResponse(QHttpServerResponder::StatusCode::BadRequest); } else if (!loadModel(modelInfo)) { - std::cerr << "ERROR: couldn't load model " << modelInfo.name.toStdString() << std::endl; + std::cerr << "ERROR: couldn't load model " << modelInfo.name().toStdString() << std::endl; return QHttpServerResponse(QHttpServerResponder::StatusCode::InternalServerError); } // don't remember any context resetContext(); - const QString promptTemplate = MySettings::globalInstance()->promptTemplate(); - const float top_k = MySettings::globalInstance()->topK(); - const int n_batch = MySettings::globalInstance()->promptBatchSize(); - const float repeat_penalty = MySettings::globalInstance()->repeatPenalty(); - const int repeat_last_n = MySettings::globalInstance()->repeatPenaltyTokens(); - - int threadCount = MySettings::globalInstance()->threadCount(); - if (threadCount <= 0) - threadCount = std::min(4, (int32_t) std::thread::hardware_concurrency()); + const QString promptTemplate = modelInfo.promptTemplate(); + const float top_k = modelInfo.topK(); + const int n_batch = modelInfo.promptBatchSize(); + const float repeat_penalty = modelInfo.repeatPenalty(); + const int repeat_last_n = modelInfo.repeatPenaltyTokens(); int promptTokens = 0; int responseTokens = 0; QList>> responses; for (int i = 0; i < n; ++i) { - if (!prompt( + if (!promptInternal( m_collections, actualPrompt, promptTemplate, @@ -319,10 +315,9 @@ QHttpServerResponse Server::handleCompletionRequest(const QHttpServerRequest &re temperature, n_batch, repeat_penalty, - repeat_last_n, - threadCount)) { + repeat_last_n)) { - std::cerr << "ERROR: couldn't prompt model " << modelInfo.name.toStdString() << std::endl; + std::cerr << "ERROR: couldn't prompt model " << modelInfo.name().toStdString() << std::endl; return QHttpServerResponse(QHttpServerResponder::StatusCode::InternalServerError); } QString echoedPrompt = actualPrompt; @@ -340,7 +335,7 @@ QHttpServerResponse Server::handleCompletionRequest(const QHttpServerRequest &re responseObject.insert("id", "foobarbaz"); responseObject.insert("object", "text_completion"); responseObject.insert("created", QDateTime::currentSecsSinceEpoch()); - responseObject.insert("model", modelInfo.name); + responseObject.insert("model", modelInfo.name()); QJsonArray choices;