diff --git a/common/common.cmake b/common/common.cmake index a60eb33af851..bf32054daa80 100644 --- a/common/common.cmake +++ b/common/common.cmake @@ -11,7 +11,6 @@ function(gpt4all_add_warning_options target) -Wextra-semi -Wformat=2 -Wmissing-include-dirs - -Wnull-dereference -Wstrict-overflow=2 -Wvla # errors @@ -22,8 +21,6 @@ function(gpt4all_add_warning_options target) # disabled warnings -Wno-sign-compare -Wno-unused-parameter - -Wno-unused-function - -Wno-unused-variable ) if (CMAKE_CXX_COMPILER_ID STREQUAL "GNU") target_compile_options("${target}" PRIVATE diff --git a/gpt4all-backend/deps/llama.cpp-mainline b/gpt4all-backend/deps/llama.cpp-mainline index b3b5c0571eda..58a55efc4ae5 160000 --- a/gpt4all-backend/deps/llama.cpp-mainline +++ b/gpt4all-backend/deps/llama.cpp-mainline @@ -1 +1 @@ -Subproject commit b3b5c0571eda3065035a7f25f7b84640b159d821 +Subproject commit 58a55efc4ae5dd3bc12887d47981faa7136027af diff --git a/gpt4all-backend/include/gpt4all-backend/llmodel.h b/gpt4all-backend/include/gpt4all-backend/llmodel.h index 1b60dae6faa2..ecb4a6ffe244 100644 --- a/gpt4all-backend/include/gpt4all-backend/llmodel.h +++ b/gpt4all-backend/include/gpt4all-backend/llmodel.h @@ -213,7 +213,7 @@ class LLModel { protected: // These are pure virtual because subclasses need to implement as the default implementation of // 'prompt' above calls these functions - virtual std::vector tokenize(PromptContext &ctx, std::string_view str, bool special = false) = 0; + virtual std::vector tokenize(std::string_view str, bool special = false) = 0; virtual bool isSpecialToken(Token id) const = 0; virtual std::string tokenToString(Token id) const = 0; virtual void initSampler(PromptContext &ctx) = 0; diff --git a/gpt4all-backend/src/llamamodel.cpp b/gpt4all-backend/src/llamamodel.cpp index cd3ffa578177..d34024f432a5 100644 --- a/gpt4all-backend/src/llamamodel.cpp +++ b/gpt4all-backend/src/llamamodel.cpp @@ -511,7 +511,7 @@ size_t LLamaModel::restoreState(std::span src) return llama_state_set_data(d_ptr->ctx, src.data(), src.size()); } -std::vector LLamaModel::tokenize(PromptContext &ctx, std::string_view str, bool special) +std::vector LLamaModel::tokenize(std::string_view str, bool special) { bool atStart = m_tokenize_last_token == -1; bool insertSpace = atStart || isSpecialToken(m_tokenize_last_token); diff --git a/gpt4all-backend/src/llamamodel_impl.h b/gpt4all-backend/src/llamamodel_impl.h index 0d0ddf34abc2..f7b3a47e0b6e 100644 --- a/gpt4all-backend/src/llamamodel_impl.h +++ b/gpt4all-backend/src/llamamodel_impl.h @@ -54,7 +54,7 @@ class LLamaModel : public LLModel { bool m_supportsCompletion = false; protected: - std::vector tokenize(PromptContext &ctx, std::string_view str, bool special) override; + std::vector tokenize(std::string_view str, bool special) override; bool isSpecialToken(Token id) const override; std::string tokenToString(Token id) const override; void initSampler(PromptContext &ctx) override; diff --git a/gpt4all-backend/src/llmodel_shared.cpp b/gpt4all-backend/src/llmodel_shared.cpp index c1e969d4b940..3868f0d07ad3 100644 --- a/gpt4all-backend/src/llmodel_shared.cpp +++ b/gpt4all-backend/src/llmodel_shared.cpp @@ -90,41 +90,33 @@ void LLModel::prompt(const std::string &prompt, } } - auto old_n_past = promptCtx.n_past; // prepare to fake n_past for tokenize - // tokenize the user prompt std::vector embd_inp; if (placeholders.empty()) { // this is unusual, but well-defined std::cerr << __func__ << ": prompt template has no placeholder\n"; - embd_inp = tokenize(promptCtx, promptTemplate, true); + embd_inp = tokenize(promptTemplate, true); } else { // template: beginning of user prompt const auto &phUser = placeholders[0]; std::string userPrefix(phUser.prefix()); - if (!userPrefix.empty()) { - embd_inp = tokenize(promptCtx, userPrefix, true); - promptCtx.n_past += embd_inp.size(); - } + if (!userPrefix.empty()) + embd_inp = tokenize(userPrefix, true); // user input (shouldn't have special token processing) - auto tokens = tokenize(promptCtx, prompt, special); + auto tokens = tokenize(prompt, special); embd_inp.insert(embd_inp.end(), tokens.begin(), tokens.end()); - promptCtx.n_past += tokens.size(); // template: end of user prompt + start of assistant prompt size_t start = phUser.position() + phUser.length(); size_t end = placeholders.size() >= 2 ? placeholders[1].position() : promptTemplate.length(); auto userToAsst = promptTemplate.substr(start, end - start); if (!userToAsst.empty()) { - tokens = tokenize(promptCtx, userToAsst, true); + tokens = tokenize(userToAsst, true); embd_inp.insert(embd_inp.end(), tokens.begin(), tokens.end()); - promptCtx.n_past += tokens.size(); } } - promptCtx.n_past = old_n_past; // restore n_past so decodePrompt can increment it - // decode the user prompt if (!decodePrompt(promptCallback, responseCallback, allowContextShift, promptCtx, embd_inp)) return; // error @@ -133,7 +125,7 @@ void LLModel::prompt(const std::string &prompt, if (!fakeReply) { generateResponse(responseCallback, allowContextShift, promptCtx); } else { - embd_inp = tokenize(promptCtx, *fakeReply, false); + embd_inp = tokenize(*fakeReply, false); if (!decodePrompt(promptCallback, responseCallback, allowContextShift, promptCtx, embd_inp, true)) return; // error } @@ -148,7 +140,7 @@ void LLModel::prompt(const std::string &prompt, asstSuffix = "\n\n"; // default to a blank link, good for e.g. Alpaca } if (!asstSuffix.empty()) { - embd_inp = tokenize(promptCtx, asstSuffix, true); + embd_inp = tokenize(asstSuffix, true); decodePrompt(promptCallback, responseCallback, allowContextShift, promptCtx, embd_inp); } } diff --git a/gpt4all-chat/deps/usearch b/gpt4all-chat/deps/usearch index 1f0618a86f9d..9e59f1036657 160000 --- a/gpt4all-chat/deps/usearch +++ b/gpt4all-chat/deps/usearch @@ -1 +1 @@ -Subproject commit 1f0618a86f9dbb7386237241cee96cc425dd7b55 +Subproject commit 9e59f1036657303b29eaf709945f339e403e5f2f diff --git a/gpt4all-chat/src/chat.cpp b/gpt4all-chat/src/chat.cpp index 347a9cd24c12..6b9c8898624f 100644 --- a/gpt4all-chat/src/chat.cpp +++ b/gpt4all-chat/src/chat.cpp @@ -31,7 +31,7 @@ Chat::Chat(QObject *parent) connectLLM(); } -Chat::Chat(bool isServer, QObject *parent) +Chat::Chat(server_tag_t, QObject *parent) : QObject(parent) , m_id(Network::globalInstance()->generateUniqueId()) , m_name(tr("Server Chat")) diff --git a/gpt4all-chat/src/chat.h b/gpt4all-chat/src/chat.h index 2d98322fef8d..05a6878d3520 100644 --- a/gpt4all-chat/src/chat.h +++ b/gpt4all-chat/src/chat.h @@ -45,6 +45,10 @@ class Chat : public QObject QML_UNCREATABLE("Only creatable from c++!") public: + // tag for constructing a server chat + struct server_tag_t { explicit server_tag_t() = default; }; + static inline constexpr server_tag_t server_tag = server_tag_t(); + enum ResponseState { ResponseStopped, LocalDocsRetrieval, @@ -56,7 +60,7 @@ class Chat : public QObject Q_ENUM(ResponseState) explicit Chat(QObject *parent = nullptr); - explicit Chat(bool isServer, QObject *parent = nullptr); + explicit Chat(server_tag_t, QObject *parent = nullptr); virtual ~Chat(); void destroy() { m_llmodel->destroy(); } void connectLLM(); diff --git a/gpt4all-chat/src/chatapi.h b/gpt4all-chat/src/chatapi.h index 463996b2302b..31d05310c6a7 100644 --- a/gpt4all-chat/src/chatapi.h +++ b/gpt4all-chat/src/chatapi.h @@ -98,9 +98,8 @@ class ChatAPI : public QObject, public LLModel { // them as they are only called from the default implementation of 'prompt' which we override and // completely replace - std::vector tokenize(PromptContext &ctx, std::string_view str, bool special) override + std::vector tokenize(std::string_view str, bool special) override { - (void)ctx; (void)str; (void)special; throw std::logic_error("not implemented"); diff --git a/gpt4all-chat/src/chatlistmodel.h b/gpt4all-chat/src/chatlistmodel.h index 95f8e6f2a523..a30efd5aff32 100644 --- a/gpt4all-chat/src/chatlistmodel.h +++ b/gpt4all-chat/src/chatlistmodel.h @@ -147,7 +147,7 @@ class ChatListModel : public QAbstractListModel if (m_serverChat) return; - m_serverChat = new Chat(true /*isServer*/, this); + m_serverChat = new Chat(Chat::server_tag, this); beginInsertRows(QModelIndex(), m_chats.size(), m_chats.size()); m_chats.append(m_serverChat); endInsertRows(); diff --git a/gpt4all-chat/src/chatllm.cpp b/gpt4all-chat/src/chatllm.cpp index 2b133ce760bf..9898ba22f49e 100644 --- a/gpt4all-chat/src/chatllm.cpp +++ b/gpt4all-chat/src/chatllm.cpp @@ -179,6 +179,8 @@ void ChatLLM::handleForceMetalChanged(bool forceMetal) reloadModel(); m_reloadingToChangeVariant = false; } +#else + Q_UNUSED(forceMetal); #endif } diff --git a/gpt4all-chat/src/chatllm.h b/gpt4all-chat/src/chatllm.h index 201c7779516c..98348f5f891a 100644 --- a/gpt4all-chat/src/chatllm.h +++ b/gpt4all-chat/src/chatllm.h @@ -51,7 +51,7 @@ enum class LLModelTypeV1 { // since chat version 6 (v2.5.0) NONE = -1, // no state }; -static LLModelTypeV1 parseLLModelTypeV1(int type) +inline LLModelTypeV1 parseLLModelTypeV1(int type) { switch (LLModelTypeV1(type)) { case LLModelTypeV1::GPTJ: @@ -68,7 +68,7 @@ static LLModelTypeV1 parseLLModelTypeV1(int type) } } -static LLModelTypeV1 parseLLModelTypeV0(int v0) +inline LLModelTypeV1 parseLLModelTypeV0(int v0) { switch (LLModelTypeV0(v0)) { case LLModelTypeV0::MPT: return LLModelTypeV1::MPT; diff --git a/gpt4all-chat/src/chatviewtextprocessor.cpp b/gpt4all-chat/src/chatviewtextprocessor.cpp index 3dba986cb449..c7c221a3d991 100644 --- a/gpt4all-chat/src/chatviewtextprocessor.cpp +++ b/gpt4all-chat/src/chatviewtextprocessor.cpp @@ -967,8 +967,6 @@ void ChatViewTextProcessor::handleCodeBlocks() cursor.setPosition(matchesCode[index].capturedEnd(), QTextCursor::KeepAnchor); cursor.removeSelectedText(); - int startPos = cursor.position(); - QTextFrameFormat frameFormat = frameFormatBase; QString capturedText = matchesCode[index].captured(1); QString codeLanguage; @@ -1004,7 +1002,7 @@ void ChatViewTextProcessor::handleCodeBlocks() QTextFrame *mainFrame = cursor.currentFrame(); cursor.setCharFormat(textFormat); - QTextFrame *frame = cursor.insertFrame(frameFormat); + cursor.insertFrame(frameFormat); QTextTable *table = cursor.insertTable(codeLanguage.isEmpty() ? 1 : 2, 1, tableFormat); if (!codeLanguage.isEmpty()) { @@ -1016,7 +1014,6 @@ void ChatViewTextProcessor::handleCodeBlocks() headerCursor.insertText(codeLanguage); QTextTableCell copy = headerTable->cellAt(0, 1); QTextCursor copyCursor = copy.firstCursorPosition(); - int startPos = copyCursor.position(); CodeCopy newCopy; newCopy.text = lines.join("\n"); newCopy.startPos = copyCursor.position(); diff --git a/gpt4all-chat/src/database.cpp b/gpt4all-chat/src/database.cpp index bec2e54810aa..1b86caace5e7 100644 --- a/gpt4all-chat/src/database.cpp +++ b/gpt4all-chat/src/database.cpp @@ -290,7 +290,7 @@ static bool selectCountChunks(QSqlQuery &q, int folder_id, int &count) return true; } -static bool selectChunk(QSqlQuery &q, const QList &chunk_ids, int retrievalSize) +static bool selectChunk(QSqlQuery &q, const QList &chunk_ids) { QString chunk_ids_str = QString::number(chunk_ids[0]); for (size_t i = 1; i < chunk_ids.size(); ++i) @@ -307,10 +307,6 @@ static const QString INSERT_COLLECTION_SQL = uR"( returning id; )"_s; -static const QString DELETE_COLLECTION_SQL = uR"( - delete from collections where name = ? and folder_id = ?; - )"_s; - static const QString SELECT_FOLDERS_FROM_COLLECTIONS_SQL = uR"( select f.id, f.path from collections c @@ -379,15 +375,6 @@ static bool addCollection(QSqlQuery &q, const QString &collection_name, const QD return true; } -static bool removeCollection(QSqlQuery &q, const QString &collection_name, int folder_id) -{ - if (!q.prepare(DELETE_COLLECTION_SQL)) - return false; - q.addBindValue(collection_name); - q.addBindValue(folder_id); - return q.exec(); -} - static bool selectFoldersFromCollection(QSqlQuery &q, const QString &collection_name, QList> *folders) { if (!q.prepare(SELECT_FOLDERS_FROM_COLLECTIONS_SQL)) @@ -520,10 +507,6 @@ static const QString GET_FOLDER_EMBEDDING_MODEL_SQL = uR"( where ci.folder_id = ?; )"_s; -static const QString SELECT_ALL_FOLDERPATHS_SQL = uR"( - select path from folders; - )"_s; - static const QString FOLDER_REMOVE_ALL_DOCS_SQL[] = { uR"( delete from embeddings @@ -598,17 +581,6 @@ static bool sqlGetFolderEmbeddingModel(QSqlQuery &q, int id, QString &embedding_ return true; } -static bool selectAllFolderPaths(QSqlQuery &q, QList *folder_paths) -{ - if (!q.prepare(SELECT_ALL_FOLDERPATHS_SQL)) - return false; - if (!q.exec()) - return false; - while (q.next()) - folder_paths->append(q.value(0).toString()); - return true; -} - static const QString INSERT_COLLECTION_ITEM_SQL = uR"( insert into collection_items(collection_id, folder_id) values(?, ?) @@ -2499,7 +2471,7 @@ void Database::retrieveFromDB(const QList &collections, const QString & return; QSqlQuery q(m_db); - if (!selectChunk(q, searchResults, retrievalSize)) { + if (!selectChunk(q, searchResults)) { qDebug() << "ERROR: selecting chunks:" << q.lastError(); return; } diff --git a/gpt4all-chat/src/download.cpp b/gpt4all-chat/src/download.cpp index e773a41fe385..0f5b7b5dd570 100644 --- a/gpt4all-chat/src/download.cpp +++ b/gpt4all-chat/src/download.cpp @@ -58,11 +58,6 @@ Download::Download() m_startTime = QDateTime::currentDateTime(); } -static bool operator==(const ReleaseInfo& lhs, const ReleaseInfo& rhs) -{ - return lhs.version == rhs.version; -} - std::strong_ordering Download::compareAppVersions(const QString &a, const QString &b) { static QRegularExpression versionRegex(R"(^(\d+(?:\.\d+){0,2})(-.+)?$)");