" << "" << "";
}
-
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: %1 %2 ";
+ }
};
} // namespace QodeAssist::Templates
diff --git a/templates/CustomFimTemplate.hpp b/templates/CustomFimTemplate.hpp
index 9180c32..5b5fd83 100644
--- a/templates/CustomFimTemplate.hpp
+++ b/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
diff --git a/templates/DeepSeekCoderFim.hpp b/templates/DeepSeekCoderFim.hpp
index d7aa98f..8bfdd0a 100644
--- a/templates/DeepSeekCoderFim.hpp
+++ b/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
diff --git a/templates/CodeLlamaChat.hpp b/templates/Llama2.hpp
similarity index 54%
rename from templates/CodeLlamaChat.hpp
rename to templates/Llama2.hpp
index 0a6b713..6888f7b 100644
--- a/templates/CodeLlamaChat.hpp
+++ b/templates/Llama2.hpp
@@ -19,38 +19,46 @@
#pragma once
-#include
-
#include "llmcore/PromptTemplate.hpp"
+#include
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 "[INST] %1 [/INST]"; }
- QStringList stopWords() const override { return QStringList() << "[INST]" << "[/INST]"; }
-
+ QString promptTemplate() const override { return {}; }
+ QStringList stopWords() const override { return QStringList() << "[INST]"; }
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);
+ 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 = QString("[INST]<>\n%1\n<>[/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;
}
-};
-
-class LlamaChat : public CodeLlamaChat
-{
-public:
- QString name() const override { return "Llama Chat"; }
+ QString description() const override
+ {
+ return "The message will contain the following tokens: [INST]%1[/INST]\n";
+ }
};
} // namespace QodeAssist::Templates
diff --git a/templates/StarCoderChat.hpp b/templates/Llama3.hpp
similarity index 59%
rename from templates/StarCoderChat.hpp
rename to templates/Llama3.hpp
index 437f53e..7b3bf2b 100644
--- a/templates/StarCoderChat.hpp
+++ b/templates/Llama3.hpp
@@ -20,32 +20,43 @@
#pragma once
#include
+
#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() << "###"
- << "<|endoftext|>" << "";
+ return QStringList() << "<|start_header_id|>" << "<|end_header_id|>" << "<|eot_id|>";
}
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);
+ for (int i = 0; i < messages.size(); ++i) {
+ QJsonObject message = messages[i].toObject();
+ QString role = message["role"].toString();
+ 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
diff --git a/templates/Ollama.hpp b/templates/Ollama.hpp
index e215c66..b434805 100644
--- a/templates/Ollama.hpp
+++ b/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
diff --git a/templates/Qwen.hpp b/templates/Qwen.hpp
index cf36629..1231a74 100644
--- a/templates/Qwen.hpp
+++ b/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
diff --git a/templates/StarCoder2Fim.hpp b/templates/StarCoder2Fim.hpp
index e0d3ca2..00f68da 100644
--- a/templates/StarCoder2Fim.hpp
+++ b/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: "
+ "%1%2";
+ }
};
} // namespace QodeAssist::Templates
diff --git a/templates/Templates.hpp b/templates/Templates.hpp
new file mode 100644
index 0000000..098aaef
--- /dev/null
+++ b/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 .
+ */
+
+#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();
+ templateManager.registerTemplate();
+ templateManager.registerTemplate();
+ templateManager.registerTemplate();
+ templateManager.registerTemplate();
+ templateManager.registerTemplate();
+ templateManager.registerTemplate();
+ templateManager.registerTemplate();
+ templateManager.registerTemplate();
+ templateManager.registerTemplate();
+ templateManager.registerTemplate();
+ templateManager.registerTemplate();
+}
+
+} // namespace QodeAssist::Templates