♻️ refactor: Rework currents and add new templates

Add Alpaca, Llama3, LLama2, ChatML templates

CMakeLists.txt

@@ -40,16 +40,22 @@ add_qtc_plugin(QodeAssist
     QodeAssistConstants.hpp
     QodeAssisttr.h
     LLMClientInterface.hpp LLMClientInterface.cpp
+    templates/Templates.hpp
     templates/CodeLlamaFim.hpp
     templates/StarCoder2Fim.hpp
     templates/DeepSeekCoderFim.hpp
     templates/CustomFimTemplate.hpp
-    templates/DeepSeekCoderChat.hpp
+
-    templates/CodeLlamaChat.hpp
+
     templates/Qwen.hpp
-    templates/StarCoderChat.hpp
+
     templates/Ollama.hpp
     templates/BasicChat.hpp
+    templates/Llama3.hpp
+    templates/ChatML.hpp
+    templates/Alpaca.hpp
+    templates/Llama2.hpp
+    providers/Providers.hpp
     providers/OllamaProvider.hpp providers/OllamaProvider.cpp
     providers/LMStudioProvider.hpp providers/LMStudioProvider.cpp
     providers/OpenAICompatProvider.hpp providers/OpenAICompatProvider.cpp

ChatView/ClientInterface.cpp

@@ -68,6 +68,8 @@ void ClientInterface::sendMessage(const QString &message, bool includeCurrentFil
 {
     cancelRequest();
 
+    m_chatModel->addMessage(message, ChatModel::ChatRole::User, "");
+
     auto &chatAssistantSettings = Settings::chatAssistantSettings();
 
     auto providerName = Settings::generalSettings().caProvider();
@@ -128,7 +130,6 @@ void ClientInterface::sendMessage(const QString &message, bool includeCurrentFil
     QJsonObject request;
     request["id"] = QUuid::createUuid().toString();
 
-    m_chatModel->addMessage(message, ChatModel::ChatRole::User, "");
     m_requestHandler->sendLLMRequest(config, request);
 }
 

llmcore/PromptTemplate.hpp

@@ -38,5 +38,6 @@ public:
     virtual QString promptTemplate() const = 0;
     virtual QStringList stopWords() const = 0;
     virtual void prepareRequest(QJsonObject &request, const ContextData &context) const = 0;
+    virtual QString description() const = 0;
 };
 } // namespace QodeAssist::LLMCore

providers/Providers.hpp

@@ -0,0 +1,37 @@
+/* 
+ * Copyright (C) 2024 Petr Mironychev
+ *
+ * This file is part of QodeAssist.
+ *
+ * QodeAssist is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation, either version 3 of the License, or
+ * (at your option) any later version.
+ *
+ * QodeAssist is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with QodeAssist. If not, see <https://www.gnu.org/licenses/>.
+ */
+
+#pragma once
+
+#include "llmcore/ProvidersManager.hpp"
+#include "providers/LMStudioProvider.hpp"
+#include "providers/OllamaProvider.hpp"
+#include "providers/OpenAICompatProvider.hpp"
+
+namespace QodeAssist::Providers {
+
+inline void registerProviders()
+{
+    auto &providerManager = LLMCore::ProvidersManager::instance();
+    providerManager.registerProvider<OllamaProvider>();
+    providerManager.registerProvider<LMStudioProvider>();
+    providerManager.registerProvider<OpenAICompatProvider>();
+}
+
+} // namespace QodeAssist::Providers

qodeassist.cpp

@@ -43,22 +43,9 @@
 #include "QodeAssistClient.hpp"
 #include "chat/ChatOutputPane.h"
 #include "chat/NavigationPanel.hpp"
-#include "llmcore/PromptTemplateManager.hpp"
-#include "llmcore/ProvidersManager.hpp"
-#include "providers/LMStudioProvider.hpp"
-#include "providers/OllamaProvider.hpp"
-#include "providers/OpenAICompatProvider.hpp"
 
-#include "templates/BasicChat.hpp"
+#include "providers/Providers.hpp"
-#include "templates/CodeLlamaChat.hpp"
+#include "templates/Templates.hpp"
-#include "templates/CodeLlamaFim.hpp"
-#include "templates/CustomFimTemplate.hpp"
-#include "templates/DeepSeekCoderChat.hpp"
-#include "templates/DeepSeekCoderFim.hpp"
-#include "templates/Ollama.hpp"
-#include "templates/Qwen.hpp"
-#include "templates/StarCoder2Fim.hpp"
-#include "templates/StarCoderChat.hpp"
 
 using namespace Utils;
 using namespace Core;
@@ -85,25 +72,8 @@ public:
 
     void initialize() final
     {
-        auto &providerManager = LLMCore::ProvidersManager::instance();
+        Providers::registerProviders();
-        providerManager.registerProvider<Providers::OllamaProvider>();
+        Templates::registerTemplates();
-        providerManager.registerProvider<Providers::LMStudioProvider>();
-        providerManager.registerProvider<Providers::OpenAICompatProvider>();
-
-        auto &templateManager = LLMCore::PromptTemplateManager::instance();
-        templateManager.registerTemplate<Templates::CodeLlamaFim>();
-        templateManager.registerTemplate<Templates::StarCoder2Fim>();
-        templateManager.registerTemplate<Templates::DeepSeekCoderFim>();
-        templateManager.registerTemplate<Templates::CustomTemplate>();
-        templateManager.registerTemplate<Templates::DeepSeekCoderChat>();
-        templateManager.registerTemplate<Templates::CodeLlamaChat>();
-        templateManager.registerTemplate<Templates::LlamaChat>();
-        templateManager.registerTemplate<Templates::StarCoderChat>();
-        templateManager.registerTemplate<Templates::QwenChat>();
-        templateManager.registerTemplate<Templates::QwenFim>();
-        templateManager.registerTemplate<Templates::OllamaAutoFim>();
-        templateManager.registerTemplate<Templates::OllamaAutoChat>();
-        templateManager.registerTemplate<Templates::BasicChat>();
 
         Utils::Icon QCODEASSIST_ICON(
             {{":/resources/images/qoderassist-icon.png", Utils::Theme::IconsBaseColor}});

templates/Alpaca.hpp

@@ -0,0 +1,67 @@
+/* 
+ * Copyright (C) 2024 Petr Mironychev
+ *
+ * This file is part of QodeAssist.
+ *
+ * QodeAssist is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation, either version 3 of the License, or
+ * (at your option) any later version.
+ *
+ * QodeAssist is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with QodeAssist. If not, see <https://www.gnu.org/licenses/>.
+ */
+
+#pragma once
+
+#include "llmcore/PromptTemplate.hpp"
+#include <QJsonArray>
+
+namespace QodeAssist::Templates {
+
+class Alpaca : public LLMCore::PromptTemplate
+{
+public:
+    QString name() const override { return "Alpaca"; }
+    LLMCore::TemplateType type() const override { return LLMCore::TemplateType::Chat; }
+    QString promptTemplate() const override { return {}; }
+    QStringList stopWords() const override
+    {
+        return QStringList() << "### Instruction:" << "### Response:";
+    }
+    void prepareRequest(QJsonObject &request, const LLMCore::ContextData &context) const override
+    {
+        QJsonArray messages = request["messages"].toArray();
+
+        for (int i = 0; i < messages.size(); ++i) {
+            QJsonObject message = messages[i].toObject();
+            QString role = message["role"].toString();
+            QString content = message["content"].toString();
+
+            QString formattedContent;
+            if (role == "system") {
+                formattedContent = content + "\n\n";
+            } else if (role == "user") {
+                formattedContent = "### Instruction:\n" + content + "\n\n";
+            } else if (role == "assistant") {
+                formattedContent = "### Response:\n" + content + "\n\n";
+            }
+
+            message["content"] = formattedContent;
+            messages[i] = message;
+        }
+
+        request["messages"] = messages;
+    }
+    QString description() const override
+    {
+        return "The message will contain the following tokens: ### Instruction:\n### Response:\n";
+    }
+};
+
+} // namespace QodeAssist::Templates

templates/BasicChat.hpp

@@ -32,18 +32,9 @@ public:
     QString name() const override { return "Basic Chat"; }
     QString promptTemplate() const override { return {}; }
     QStringList stopWords() const override { return QStringList(); }
-
     void prepareRequest(QJsonObject &request, const LLMCore::ContextData &context) const override
-    {
+    {}
-        QJsonArray messages = request["messages"].toArray();
+    QString description() const override { return "chat without tokens"; }
-
-        QJsonObject newMessage;
-        newMessage["role"] = "user";
-        newMessage["content"] = context.prefix;
-        messages.append(newMessage);
-
-        request["messages"] = messages;
-    }
 };
 
 } // namespace QodeAssist::Templates

templates/ChatML.hpp

@@ -20,35 +20,41 @@
 #pragma once
 
 #include <QJsonArray>
+
 #include "llmcore/PromptTemplate.hpp"
 
 namespace QodeAssist::Templates {
 
-class DeepSeekCoderChat : public LLMCore::PromptTemplate
+class ChatML : public LLMCore::PromptTemplate
 {
 public:
-    QString name() const override { return "DeepSeekCoder Chat"; }
+    QString name() const override { return "ChatML"; }
     LLMCore::TemplateType type() const override { return LLMCore::TemplateType::Chat; }
-
+    QString promptTemplate() const override { return {}; }
-    QString promptTemplate() const override { return "### Instruction:\n%1\n### Response:\n"; }
-
     QStringList stopWords() const override
     {
-        return QStringList() << "### Instruction:" << "### Response:" << "\n\n### " << "<|EOT|>";
+        return QStringList() << "<|im_start|>" << "<|im_end|>";
     }
-
     void prepareRequest(QJsonObject &request, const LLMCore::ContextData &context) const override
     {
-        QString formattedPrompt = promptTemplate().arg(context.prefix);
         QJsonArray messages = request["messages"].toArray();
 
-        QJsonObject newMessage;
+        for (int i = 0; i < messages.size(); ++i) {
-        newMessage["role"] = "user";
+            QJsonObject message = messages[i].toObject();
-        newMessage["content"] = formattedPrompt;
+            QString role = message["role"].toString();
-        messages.append(newMessage);
+            QString content = message["content"].toString();
+
+            message["content"] = QString("<|im_start|>%1\n%2\n<|im_end|>").arg(role, content);
+
+            messages[i] = message;
+        }
 
         request["messages"] = messages;
     }
+    QString description() const override
+    {
+        return "The message will contain the following tokens: <|im_start|>%1\n%2\n<|im_end|>";
+    }
 };
 
 } // namespace QodeAssist::Templates

templates/CodeLlamaFim.hpp

@@ -33,12 +33,15 @@ public:
     {
         return QStringList() << "<EOT>" << "<PRE>" << "<SUF" << "<MID>";
     }
-
     void prepareRequest(QJsonObject &request, const LLMCore::ContextData &context) const override
     {
         QString formattedPrompt = promptTemplate().arg(context.prefix, context.suffix);
         request["prompt"] = formattedPrompt;
     }
+    QString description() const override
+    {
+        return "The message will contain the following tokens: <PRE> %1 <SUF>%2 <MID>";
+    }
 };
 
 } // namespace QodeAssist::Templates

templates/CustomFimTemplate.hpp

@@ -39,7 +39,6 @@ public:
         return Settings::customPromptSettings().customJsonTemplate();
     }
     QStringList stopWords() const override { return QStringList(); }
-
     void prepareRequest(QJsonObject &request, const LLMCore::ContextData &context) const override
     {
         QJsonDocument doc = QJsonDocument::fromJson(promptTemplate().toUtf8());
@@ -56,6 +55,7 @@ public:
             request[it.key()] = it.value();
         }
     }
+    QString description() const override { return promptTemplate(); }
 
 private:
     QJsonValue processJsonValue(const QJsonValue &value, const LLMCore::ContextData &context) const

templates/DeepSeekCoderFim.hpp

@@ -38,6 +38,11 @@ public:
         QString formattedPrompt = promptTemplate().arg(context.prefix, context.suffix);
         request["prompt"] = formattedPrompt;
     }
+    QString description() const override
+    {
+        return "The message will contain the following tokens: "
+               "<｜fim▁begin｜>%1<｜fim▁hole｜>%2<｜fim▁end｜>";
+    }
 };
 
 } // namespace QodeAssist::Templates

templates/Llama2.hpp

@@ -19,38 +19,46 @@
 
 #pragma once
 
-#include <QJsonArray>
-
 #include "llmcore/PromptTemplate.hpp"
+#include <QJsonArray>
 
 namespace QodeAssist::Templates {
 
-class CodeLlamaChat : public LLMCore::PromptTemplate
+class Llama2 : public LLMCore::PromptTemplate
 {
 public:
+    QString name() const override { return "Llama 2"; }
     LLMCore::TemplateType type() const override { return LLMCore::TemplateType::Chat; }
-    QString name() const override { return "CodeLlama Chat"; }
+    QString promptTemplate() const override { return {}; }
-    QString promptTemplate() const override { return "[INST] %1 [/INST]"; }
+    QStringList stopWords() const override { return QStringList() << "[INST]"; }
-    QStringList stopWords() const override { return QStringList() << "[INST]" << "[/INST]"; }
-
     void prepareRequest(QJsonObject &request, const LLMCore::ContextData &context) const override
     {
-        QString formattedPrompt = promptTemplate().arg(context.prefix);
         QJsonArray messages = request["messages"].toArray();
 
-        QJsonObject newMessage;
+        for (int i = 0; i < messages.size(); ++i) {
-        newMessage["role"] = "user";
+            QJsonObject message = messages[i].toObject();
-        newMessage["content"] = formattedPrompt;
+            QString role = message["role"].toString();
-        messages.append(newMessage);
+            QString content = message["content"].toString();
+
+            QString formattedContent;
+            if (role == "system") {
+                formattedContent = QString("[INST]<<SYS>>\n%1\n<</SYS>>[/INST]\n").arg(content);
+            } else if (role == "user") {
+                formattedContent = QString("[INST]%1[/INST]\n").arg(content);
+            } else if (role == "assistant") {
+                formattedContent = content + "\n";
+            }
+
+            message["content"] = formattedContent;
+            messages[i] = message;
+        }
 
         request["messages"] = messages;
     }
-};
+    QString description() const override
-
+    {
-class LlamaChat : public CodeLlamaChat
+        return "The message will contain the following tokens: [INST]%1[/INST]\n";
-{
+    }
-public:
-    QString name() const override { return "Llama Chat"; }
 };
 
 } // namespace QodeAssist::Templates

templates/Llama3.hpp

@@ -20,32 +20,43 @@
 #pragma once
 
 #include <QJsonArray>
+
 #include "llmcore/PromptTemplate.hpp"
 
 namespace QodeAssist::Templates {
 
-class StarCoderChat : public LLMCore::PromptTemplate
+class Llama3 : public LLMCore::PromptTemplate
 {
 public:
-    QString name() const override { return "StarCoder Chat"; }
+    QString name() const override { return "Llama 3"; }
     LLMCore::TemplateType type() const override { return LLMCore::TemplateType::Chat; }
-    QString promptTemplate() const override { return "### Instruction:\n%1\n### Response:\n"; }
+    QString promptTemplate() const override { return ""; }
     QStringList stopWords() const override
     {
-        return QStringList() << "###"
+        return QStringList() << "<|start_header_id|>" << "<|end_header_id|>" << "<|eot_id|>";
-                             << "<|endoftext|>" << "<file_sep>";
     }
     void prepareRequest(QJsonObject &request, const LLMCore::ContextData &context) const override
     {
-        QString formattedPrompt = promptTemplate().arg(context.prefix);
         QJsonArray messages = request["messages"].toArray();
 
-        QJsonObject newMessage;
+        for (int i = 0; i < messages.size(); ++i) {
-        newMessage["role"] = "user";
+            QJsonObject message = messages[i].toObject();
-        newMessage["content"] = formattedPrompt;
+            QString role = message["role"].toString();
-        messages.append(newMessage);
+            QString content = message["content"].toString();
+
+            message["content"]
+                = QString("<|start_header_id|>%1<|end_header_id|>%2<|eot_id|>").arg(role, content);
+
+            messages[i] = message;
+        }
 
         request["messages"] = messages;
     }
+    QString description() const override
+    {
+        return "The message will contain the following tokens: "
+               "<|start_header_id|>%1<|end_header_id|>%2<|eot_id|>";
+    }
 };
+
 } // namespace QodeAssist::Templates

templates/Ollama.hpp

@@ -32,12 +32,12 @@ public:
     QString name() const override { return "Ollama Auto FIM"; }
     QString promptTemplate() const override { return {}; }
     QStringList stopWords() const override { return QStringList(); }
-
     void prepareRequest(QJsonObject &request, const LLMCore::ContextData &context) const override
     {
         request["prompt"] = context.prefix;
         request["suffix"] = context.suffix;
     }
+    QString description() const override { return "template will take from ollama modelfile"; }
 };
 
 class OllamaAutoChat : public LLMCore::PromptTemplate
@@ -59,6 +59,7 @@ public:
 
         request["messages"] = messages;
     }
+    QString description() const override { return "template will take from ollama modelfile"; }
 };
 
 } // namespace QodeAssist::Templates

templates/Qwen.hpp

@@ -24,33 +24,6 @@
 
 namespace QodeAssist::Templates {
 
-class QwenChat : public LLMCore::PromptTemplate
-{
-public:
-    QString name() const override { return "Qwen Chat"; }
-    LLMCore::TemplateType type() const override { return LLMCore::TemplateType::Chat; }
-
-    QString promptTemplate() const override { return "### Instruction:\n%1\n### Response:\n"; }
-
-    QStringList stopWords() const override
-    {
-        return QStringList() << "### Instruction:" << "### Response:" << "\n\n### " << "<|EOT|>";
-    }
-
-    void prepareRequest(QJsonObject &request, const LLMCore::ContextData &context) const override
-    {
-        QString formattedPrompt = promptTemplate().arg(context.prefix);
-        QJsonArray messages = request["messages"].toArray();
-
-        QJsonObject newMessage;
-        newMessage["role"] = "user";
-        newMessage["content"] = formattedPrompt;
-        messages.append(newMessage);
-
-        request["messages"] = messages;
-    }
-};
-
 class QwenFim : public LLMCore::PromptTemplate
 {
 public:
@@ -66,6 +39,11 @@ public:
         QString formattedPrompt = promptTemplate().arg(context.prefix, context.suffix);
         request["prompt"] = formattedPrompt;
     }
+    QString description() const override
+    {
+        return "The message will contain the following tokens: "
+               "<|fim_prefix|>%1<|fim_suffix|>%2<|fim_middle|>";
+    }
 };
 
 } // namespace QodeAssist::Templates

templates/StarCoder2Fim.hpp

@@ -39,6 +39,11 @@ public:
         QString formattedPrompt = promptTemplate().arg(context.prefix, context.suffix);
         request["prompt"] = formattedPrompt;
     }
+    QString description() const override
+    {
+        return "The message will contain the following tokens: "
+               "<fim_prefix>%1<fim_suffix>%2<fim_middle>";
+    }
 };
 
 } // namespace QodeAssist::Templates

templates/Templates.hpp

@@ -0,0 +1,54 @@
+/* 
+ * Copyright (C) 2024 Petr Mironychev
+ *
+ * This file is part of QodeAssist.
+ *
+ * QodeAssist is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation, either version 3 of the License, or
+ * (at your option) any later version.
+ *
+ * QodeAssist is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with QodeAssist. If not, see <https://www.gnu.org/licenses/>.
+ */
+
+#pragma once
+
+#include "llmcore/PromptTemplateManager.hpp"
+#include "templates/Alpaca.hpp"
+#include "templates/BasicChat.hpp"
+#include "templates/ChatML.hpp"
+#include "templates/CodeLlamaFim.hpp"
+#include "templates/CustomFimTemplate.hpp"
+#include "templates/DeepSeekCoderFim.hpp"
+#include "templates/Llama2.hpp"
+#include "templates/Llama3.hpp"
+#include "templates/Ollama.hpp"
+#include "templates/Qwen.hpp"
+#include "templates/StarCoder2Fim.hpp"
+
+namespace QodeAssist::Templates {
+
+inline void registerTemplates()
+{
+    auto &templateManager = LLMCore::PromptTemplateManager::instance();
+    templateManager.registerTemplate<CodeLlamaFim>();
+    templateManager.registerTemplate<StarCoder2Fim>();
+    templateManager.registerTemplate<DeepSeekCoderFim>();
+    templateManager.registerTemplate<CustomTemplate>();
+    templateManager.registerTemplate<QwenFim>();
+    templateManager.registerTemplate<OllamaAutoFim>();
+    templateManager.registerTemplate<OllamaAutoChat>();
+    templateManager.registerTemplate<BasicChat>();
+    templateManager.registerTemplate<Llama3>();
+    templateManager.registerTemplate<ChatML>();
+    templateManager.registerTemplate<Alpaca>();
+    templateManager.registerTemplate<Llama2>();
+}
+
+} // namespace QodeAssist::Templates