nodes replaces by lifecycles nodes

llama_cpp_vendor/CMakeLists.txt

@@ -11,7 +11,7 @@ find_package(ament_cmake REQUIRED)
 FetchContent_Declare(
   llama
   GIT_REPOSITORY https://github.com/ggerganov/llama.cpp.git
-  GIT_TAG        b3604
+  GIT_TAG        b3609
 )
 
 FetchContent_MakeAvailable(llama)

llama_ros/CMakeLists.txt

@@ -9,6 +9,7 @@ endif()
 find_package(ament_cmake REQUIRED)
 find_package(rclcpp REQUIRED)
 find_package(rclcpp_action REQUIRED)
+find_package(rclcpp_lifecycle REQUIRED)
 find_package(llama_msgs REQUIRED)
 find_package(llama_cpp_vendor REQUIRED)
 find_package(Threads REQUIRED)
@@ -56,6 +57,7 @@ target_link_libraries(llama_node
 ament_target_dependencies(llama_node
   rclcpp
   rclcpp_action
+  rclcpp_lifecycle
   llama_msgs
   llama_cpp_vendor
 )
@@ -78,6 +80,7 @@ target_link_libraries(llava_node
 ament_target_dependencies(llava_node
   rclcpp
   rclcpp_action
+  rclcpp_lifecycle
   llama_msgs
   llama_cpp_vendor
   cv_bridge

llama_ros/include/llama_ros/llama.hpp

@@ -142,7 +142,7 @@ class Llama {
 
   std::vector<token_prob> get_probs();
   struct completion_output sample();
-  void update_sampling_params(const struct llama_sampling_params &params);
+  void update_sampling_context(const struct llama_sampling_params &params);
 
 private:
   // lock

llama_ros/include/llama_ros/llama_node.hpp

@@ -25,6 +25,7 @@
 
 #include <rclcpp/rclcpp.hpp>
 #include <rclcpp_action/rclcpp_action.hpp>
+#include <rclcpp_lifecycle/lifecycle_node.hpp>
 
 #include <memory>
 #include <string>
@@ -39,20 +40,37 @@
 
 namespace llama_ros {
 
-class LlamaNode : public rclcpp::Node {
+using CallbackReturn =
+    rclcpp_lifecycle::node_interfaces::LifecycleNodeInterface::CallbackReturn;
+
+class LlamaNode : public rclcpp_lifecycle::LifecycleNode {
 
   using GenerateResponse = llama_msgs::action::GenerateResponse;
   using GoalHandleGenerateResponse =
       rclcpp_action::ServerGoalHandle<GenerateResponse>;
 
 public:
-  LlamaNode(bool load_llama = true);
+  LlamaNode();
+
+  rclcpp_lifecycle::node_interfaces::LifecycleNodeInterface::CallbackReturn
+  on_configure(const rclcpp_lifecycle::State &);
+  rclcpp_lifecycle::node_interfaces::LifecycleNodeInterface::CallbackReturn
+  on_activate(const rclcpp_lifecycle::State &);
+  rclcpp_lifecycle::node_interfaces::LifecycleNodeInterface::CallbackReturn
+  on_deactivate(const rclcpp_lifecycle::State &);
+  rclcpp_lifecycle::node_interfaces::LifecycleNodeInterface::CallbackReturn
+  on_cleanup(const rclcpp_lifecycle::State &);
+  rclcpp_lifecycle::node_interfaces::LifecycleNodeInterface::CallbackReturn
+  on_shutdown(const rclcpp_lifecycle::State &);
 
 protected:
-  std::shared_ptr<Llama> llama;
-  llama_utils::GptParams gpt_params;
+  std::unique_ptr<Llama> llama;
+  std::unique_ptr<llama_utils::GptParams> gpt_params;
   std::shared_ptr<GoalHandleGenerateResponse> goal_handle_;
 
+  virtual void create_llama();
+  void destroy_llama();
+
   virtual bool goal_empty(std::shared_ptr<const GenerateResponse::Goal> goal);
   virtual void
   execute(const std::shared_ptr<GoalHandleGenerateResponse> goal_handle);

llama_ros/include/llama_utils/gpt_params.hpp

@@ -25,6 +25,7 @@
 
 #include <memory>
 #include <rclcpp/rclcpp.hpp>
+#include <rclcpp_lifecycle/lifecycle_node.hpp>
 
 #include "common.h"
 #include "llama.h"
@@ -36,12 +37,17 @@ namespace llama_utils {
 class GptParams {
 
 public:
-  GptParams();
-  std::shared_ptr<struct gpt_params> load_params(rclcpp::Node *node);
+  GptParams(rclcpp_lifecycle::LifecycleNode::SharedPtr node);
+
+  void declare_params();
+  std::shared_ptr<struct gpt_params> get_params();
   bool
   update_sampling_params(const llama_msgs::msg::SamplingConfig &sampling_config,
                          int n_vocab, llama_token token_eos);
 
+  rclcpp_lifecycle::LifecycleNode::SharedPtr node;
+
+  // params
   bool debug;
   struct llava_ros::llava_params llava_params;
   std::shared_ptr<struct gpt_params> params;

llama_ros/include/llava_ros/llava_node.hpp

@@ -49,8 +49,7 @@ class LlavaNode : public llama_ros::LlamaNode {
                             bool url = false);
 
 protected:
-  std::shared_ptr<Llava> llava;
-
+  void create_llama();
   bool goal_empty(std::shared_ptr<const GenerateResponse::Goal> goal) override;
   void execute(
       const std::shared_ptr<GoalHandleGenerateResponse> goal_handle) override;

llama_ros/package.xml

@@ -14,6 +14,7 @@
 
   <depend>rclcpp</depend>
   <depend>rclcpp_action</depend>
+  <depend>rclcpp_lifecycle</depend>
   <depend>cv_bridge</depend>
   <depend>llama_msgs</depend>
   <depend>llama_cpp_vendor</depend>

llama_ros/src/llama_main.cpp

@@ -45,7 +45,15 @@ int main(int argc, char *argv[]) {
   sigaction(SIGINT, &sigint_action, NULL);
 
   rclcpp::init(argc, argv);
-  rclcpp::spin(std::make_shared<LlamaNode>());
+
+  auto node = std::make_shared<LlamaNode>();
+  node->configure();
+  node->activate();
+
+  rclcpp::executors::SingleThreadedExecutor executor;
+  executor.add_node(node->get_node_base_interface());
+  executor.spin();
+
   rclcpp::shutdown();
   return 0;
 }

llama_ros/src/llama_ros/llama.cpp

@@ -273,7 +273,7 @@ response_output Llama::generate_response(const std::string &input_prompt,
   llama_set_embeddings(this->ctx, false);
 
   // load params
-  this->update_sampling_params(this->params->sparams);
+  this->update_sampling_context(this->params->sparams);
 
   // load prompt
   this->load_prompt(input_prompt, true, true);
@@ -671,7 +671,8 @@ struct completion_output Llama::sample() {
   return result;
 }
 
-void Llama::update_sampling_params(const struct llama_sampling_params &params) {
+void Llama::update_sampling_context(
+    const struct llama_sampling_params &params) {
 
   this->ctx_sampling->params = params;
 

llama_ros/src/llama_ros/llama_node.cpp

@@ -35,13 +35,52 @@ using namespace llama_ros;
 using std::placeholders::_1;
 using std::placeholders::_2;
 
-LlamaNode::LlamaNode(bool load_llama) : rclcpp::Node("llama_node") {
+LlamaNode::LlamaNode()
+    : rclcpp_lifecycle::LifecycleNode("llama_node"), gpt_params(nullptr) {
+  RCLCPP_INFO(this->get_logger(), "%s started", this->get_name());
+}
+
+void LlamaNode::create_llama() {
+  this->llama = std::make_unique<Llama>(this->gpt_params->params,
+                                        this->gpt_params->debug);
+}
+
+void LlamaNode::destroy_llama() {
+  this->llama.reset();
+  this->llama = nullptr;
+}
 
-  if (load_llama) {
-    this->llama = std::make_shared<Llama>(this->gpt_params.load_params(this),
-                                          this->gpt_params.debug);
+/*
+*****************************
+*         LIFECYCLE         *
+*****************************
+*/
+rclcpp_lifecycle::node_interfaces::LifecycleNodeInterface::CallbackReturn
+LlamaNode::on_configure(const rclcpp_lifecycle::State &) {
+
+  RCLCPP_INFO(get_logger(), "[%s] Configuring...", this->get_name());
+
+  if (this->gpt_params == nullptr) {
+    this->gpt_params =
+        std::make_unique<llama_utils::GptParams>(this->shared_from_this());
+    this->gpt_params->declare_params();
   }
 
+  this->gpt_params->get_params();
+  RCLCPP_INFO(get_logger(), "[%s] Configured", this->get_name());
+
+  return rclcpp_lifecycle::node_interfaces::LifecycleNodeInterface::
+      CallbackReturn::SUCCESS;
+}
+
+rclcpp_lifecycle::node_interfaces::LifecycleNodeInterface::CallbackReturn
+LlamaNode::on_activate(const rclcpp_lifecycle::State &) {
+
+  RCLCPP_INFO(get_logger(), "[%s] Activating...", this->get_name());
+
+  // create llama
+  this->create_llama();
+
   // services
   this->tokenize_service_ = this->create_service<llama_msgs::srv::Tokenize>(
       "tokenize",
@@ -61,7 +100,53 @@ LlamaNode::LlamaNode(bool load_llama) : rclcpp::Node("llama_node") {
           std::bind(&LlamaNode::handle_cancel, this, _1),
           std::bind(&LlamaNode::handle_accepted, this, _1));
 
-  RCLCPP_INFO(this->get_logger(), "%s started", this->get_name());
+  RCLCPP_INFO(get_logger(), "[%s] Activated", this->get_name());
+
+  return rclcpp_lifecycle::node_interfaces::LifecycleNodeInterface::
+      CallbackReturn::SUCCESS;
+}
+
+rclcpp_lifecycle::node_interfaces::LifecycleNodeInterface::CallbackReturn
+LlamaNode::on_deactivate(const rclcpp_lifecycle::State &) {
+
+  RCLCPP_INFO(get_logger(), "[%s] Deactivating...", this->get_name());
+
+  this->destroy_llama();
+
+  this->tokenize_service_.reset();
+  this->tokenize_service_ = nullptr;
+
+  this->generate_embeddings_service_.reset();
+  this->generate_embeddings_service_ = nullptr;
+
+  this->goal_handle_ = nullptr;
+  this->generate_response_action_server_.reset();
+  this->generate_response_action_server_ = nullptr;
+
+  RCLCPP_INFO(get_logger(), "[%s] Deactivated", this->get_name());
+
+  return rclcpp_lifecycle::node_interfaces::LifecycleNodeInterface::
+      CallbackReturn::SUCCESS;
+}
+
+rclcpp_lifecycle::node_interfaces::LifecycleNodeInterface::CallbackReturn
+LlamaNode::on_cleanup(const rclcpp_lifecycle::State &) {
+
+  RCLCPP_INFO(get_logger(), "[%s] Cleaning up...", this->get_name());
+  RCLCPP_INFO(get_logger(), "[%s] Cleaned up", this->get_name());
+
+  return rclcpp_lifecycle::node_interfaces::LifecycleNodeInterface::
+      CallbackReturn::SUCCESS;
+}
+
+rclcpp_lifecycle::node_interfaces::LifecycleNodeInterface::CallbackReturn
+LlamaNode::on_shutdown(const rclcpp_lifecycle::State &) {
+
+  RCLCPP_INFO(get_logger(), "[%s] Shutting down...", this->get_name());
+  RCLCPP_INFO(get_logger(), "[%s] Shutted down", this->get_name());
+
+  return rclcpp_lifecycle::node_interfaces::LifecycleNodeInterface::
+      CallbackReturn::SUCCESS;
 }
 
 /*
@@ -78,7 +163,7 @@ void LlamaNode::tokenize_service_callback(
 
 /*
 *****************************
-*    EMBEEDINGS SERVICE     *
+*    EMBEDDINGS SERVICE     *
 *****************************
 */
 void LlamaNode::generate_embeddings_service_callback(
@@ -144,7 +229,7 @@ void LlamaNode::execute(
     return;
   }
 
-  if (this->gpt_params.debug) {
+  if (this->gpt_params->debug) {
     RCLCPP_INFO(this->get_logger(), "Prompt received:\n%s", prompt.c_str());
   }
 
@@ -155,9 +240,9 @@ void LlamaNode::execute(
 
   // update sampling params of gpt_params
   auto sampling_config = goal_handle->get_goal()->sampling_config;
-  this->gpt_params.update_sampling_params(sampling_config,
-                                          this->llama->get_n_vocab(),
-                                          this->llama->get_token_eos());
+  this->gpt_params->update_sampling_params(sampling_config,
+                                           this->llama->get_n_vocab(),
+                                           this->llama->get_token_eos());
 
   // call llama
   struct response_output output = this->llama->generate_response(