Adapted LLM model to single .tgz file to download

sparky-game · Oct 26, 2024 · 5e1752b · 5e1752b
1 parent 3f7a7ba
commit 5e1752b
Show file tree

Hide file tree

Showing 2 changed files with 16 additions and 20 deletions.
diff --git a/ai/CMakeLists.txt b/ai/CMakeLists.txt
@@ -16,6 +16,7 @@ set(SPM_ENABLE_SHARED OFF CACHE BOOL "" FORCE)
 set(GEMMA_ENABLE_TESTS OFF CACHE BOOL "" FORCE)
 set(HWY_ENABLE_EXAMPLES OFF CACHE BOOL "" FORCE)
 set(BENCHMARK_ENABLE_TESTING OFF CACHE BOOL "" FORCE)
+set(CMAKE_BUILD_TYPE "Release" CACHE STRING "" FORCE)
 
 FetchContent_Declare(
   gemma.cpp
@@ -25,22 +26,14 @@ FetchContent_Declare(
   GIT_SHALLOW    TRUE
 )
 FetchContent_Declare(
-  llm-model-weights
-  URL                 "https://huggingface.co/sparky-game/carbon/resolve/main/weights.sbs"
-  URL_HASH            SHA256=4703b49c4e7177a949a5e60d91a5078f81e0d9ce80f0afa4ab4cb9af44fd334c
-  DOWNLOAD_DIR        ${CMAKE_CURRENT_BINARY_DIR}
-  DOWNLOAD_NO_EXTRACT TRUE
-)
-FetchContent_Declare(
-  llm-model-tokenizer
-  URL                 "https://huggingface.co/sparky-game/carbon/resolve/main/tokenizer.spm"
-  URL_HASH            SHA256=61a7b147390c64585d6c3543dd6fc636906c9af3865a5548f27f31aee1d4c8e2
-  DOWNLOAD_DIR        ${CMAKE_CURRENT_BINARY_DIR}
-  DOWNLOAD_NO_EXTRACT TRUE
+  llm-model
+  URL                        "https://huggingface.co/sparky-game/carbon/resolve/main/model.tgz"
+  URL_HASH                   SHA256=6c42d35e1707910c5fd57104f85ff7e16022884deea7160dbe7c6c9c93431556
+  DOWNLOAD_DIR               ${CMAKE_CURRENT_BINARY_DIR}
+  DOWNLOAD_EXTRACT_TIMESTAMP TRUE
 )
 FetchContent_MakeAvailable(gemma.cpp)
-FetchContent_MakeAvailable(llm-model-weights)
-FetchContent_MakeAvailable(llm-model-tokenizer)
+FetchContent_MakeAvailable(llm-model)
 
 set_target_properties(
   gemma
@@ -55,4 +48,5 @@ set_target_properties(
 target_compile_options(sentencepiece-static PRIVATE "-w")
 
 add_executable(testgen testgen.cc)
+target_link_options(testgen PRIVATE "-static")
 target_link_libraries(testgen libgemma)
diff --git a/ai/testgen.cc b/ai/testgen.cc
@@ -14,6 +14,8 @@
 #define CARBON_AI_ABORT_ERROR(err) HWY_ABORT("%s:%u :: %s", __FILE__, __LINE__, (err))
 #define CARBON_AI_ABORT_USAGE      HWY_ABORT("usage: %s [-t <N>] --src <FILE> --test <FILE>", argv[0])
 
+#define LLM_MODEL_PATH "build/_deps/llm-model-src"
+
 static constexpr auto system_prompt {
   "Strictly follow the following instructions and rules:\n"
   "- Just write the function that has been requested, no main, no examples, no nonsense.\n"
@@ -37,15 +39,15 @@ static constexpr auto request_prompt {
   "Write an extended version of the unit test translation unit that includes additional unit tests that will increase the test coverage of the code under test.\n"
 };
 
-static inline std::string load_file_contents(const std::string &filepath) {
+static std::string load_file_contents(const std::string &filepath) {
   std::ifstream ifs { filepath };
   if (not ifs) throw std::runtime_error { "unable to open file for reading (`" + filepath + "`)" };
   std::stringstream ss;
   ss << "```\n" << ifs.rdbuf() << "\n```\n";
   return ss.str();
 }
 
-static inline std::vector<int> tokenize(const gcpp::GemmaTokenizer &tokenizer, const std::string &test_code, const std::string &src_code) {
+static std::vector<int> tokenize(const gcpp::GemmaTokenizer &tokenizer, const std::string &test_code, const std::string &src_code) {
   std::string prompt {
     std::string(system_prompt) + "\n" +
     test_code_prompt + "\n" +
@@ -60,7 +62,7 @@ static inline std::vector<int> tokenize(const gcpp::GemmaTokenizer &tokenizer, c
   return tokens;
 }
 
-static inline void preprocess_output(std::stringstream &ss) {
+static void preprocess_output(std::stringstream &ss) {
   std::string line;
   std::vector<std::string> lines;
   while (std::getline(ss, line)) lines.emplace_back(line);
@@ -71,7 +73,7 @@ static inline void preprocess_output(std::stringstream &ss) {
   }
 }
 
-static inline bool it_builds(const std::stringstream &ss) {
+static bool it_builds(const std::stringstream &ss) {
   std::ofstream ofs { "tmp.cc" };
   if (not ofs) throw std::runtime_error { "unable to open/create file for writing (`./tmp.cc`)" };
   ofs << ss.str();
@@ -84,8 +86,8 @@ static inline bool it_builds(const std::stringstream &ss) {
 
 int main(int argc, char **argv) {
   gcpp::LoaderArgs loader(argc, argv);
-  loader.tokenizer = "build/tokenizer.spm";
-  loader.weights = "build/weights.sbs";
+  loader.tokenizer = LLM_MODEL_PATH "tokenizer.spm";
+  loader.weights = LLM_MODEL_PATH "weights.sbs";
   loader.model_type_str = "7b-it";
   loader.weight_type_str = "sfp";