Chat Template

class ChatTemplate

Public Functions

ChatTemplate()

~ChatTemplate()

ChatTemplate(ChatTemplate&&) noexcept

ChatTemplate &operator=(ChatTemplate&&) noexcept

ChatTemplate(ChatTemplate const&) = delete

ChatTemplate &operator=(ChatTemplate const&) = delete

bool load(

std::filesystem::path const &modelDir,

std::string bosToken = {},

std::string eosToken = {}

)

Load the provider Jinja templates or an explicit native-renderer marker.

bool apply(

rt::LLMGenerationRequest::Request const &request,

rt::LLMGenerationRequest::FormattedRequest &formattedRequest,

Options const &options

) const

Render one structured conversation.

bool isLoaded() const noexcept

Public Static Functions

static Options optionsFrom(rt::LLMGenerationRequest const &request)

Translate the runtime request controls into renderer options.

class Impl

Public Functions

inline bool load(

std::filesystem::path const &modelDir,

std::string bosToken,

std::string eosToken

)

inline bool apply(

rt::LLMGenerationRequest::Request const &request,

rt::LLMGenerationRequest::FormattedRequest &formattedRequest,

ChatTemplate::Options const &options

) const

inline bool isLoaded() const noexcept

struct Options

Public Members

bool applyTemplate = {true}

bool addGenerationPrompt = {true}

bool enableThinking = {false}

std::string reasoningEffort

std::vector<rt::ToolDefinition> tools

rt::ToolChoice toolChoice

bool parallelToolCalls = {true}

struct Options

Public Members

bool applyTemplate = {true}

bool addGenerationPrompt = {true}

bool enableThinking = {false}

std::string reasoningEffort

std::vector<rt::ToolDefinition> tools

rt::ToolChoice toolChoice

bool parallelToolCalls = {true}