llama.cpp updated

mgonzs13 · Jan 5, 2025 · b9e5d18 · b9e5d18
1 parent f0a0af9
commit b9e5d18
Show file tree

Hide file tree

Showing 4 changed files with 9 additions and 7 deletions.
diff --git a/llama_cpp_vendor/CMakeLists.txt b/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        b4393
+  GIT_TAG        b4418
 )
 
 option(LLAMA_BUILD_COMMON "llama: build common utils library" ON)

diff --git a/llama_ros/include/llama_ros/llama.hpp b/llama_ros/include/llama_ros/llama.hpp
@@ -233,9 +233,10 @@ class Llama {
   struct common_params params;
 
   // model
+  struct common_init_result llama_init;
   struct llama_context *ctx;
   struct llama_model *model;
-  std::vector<struct common_lora_adapter_container> lora_adapters;
+  std::vector<common_lora_adapter_info> lora_adapters;
   struct common_sampler *sampler;
   struct ggml_threadpool *threadpool;
   struct ggml_threadpool *threadpool_batch;

diff --git a/llama_ros/src/llama_ros/llama.cpp b/llama_ros/src/llama_ros/llama.cpp
@@ -43,10 +43,10 @@ Llama::Llama(const struct common_params &params, std::string system_prompt)
   llama_backend_init();
   llama_numa_init(this->params.numa);
 
-  struct common_init_result llama_init = common_init_from_params(this->params);
-  this->model = llama_init.model;
-  this->ctx = llama_init.context;
-  this->lora_adapters = llama_init.lora_adapters;
+  this->llama_init = common_init_from_params(this->params);
+  this->model = llama_init.model.get();
+  this->ctx = llama_init.context.get();
+  this->lora_adapters = this->params.lora_adapters;
 
   if (this->model == NULL) {
     LLAMA_LOG_ERROR("Unable to load model");

diff --git a/llama_ros/src/llama_utils/llama_params.cpp b/llama_ros/src/llama_utils/llama_params.cpp
@@ -310,7 +310,8 @@ struct LlamaParams llama_utils::get_llama_params(
           scale = 1.0;
         }
 
-        params.params.lora_adapters.push_back({lora_adapters.at(i), scale});
+        params.params.lora_adapters.push_back(
+            {lora_adapters.at(i), scale, nullptr});
       }
     }
   }