LLM inference in C/C++
C++ ggml
tests : fix ggml init (#29554)
Georgi Gerganov81ef10ea58fc
diff --git a/include/llama.h b/include/llama.hindex ce454df527..e808fa9ec1 100644--- a/include/llama.h+++ b/include/llama.h@@ -482,14 +482,13 @@ extern "C" {LLAMA_API struct llama_model_quantize_params llama_model_quantize_default_params(void);// Initialize the llama + ggml backend- // If numa is true, use NUMA optimizations// Call once at the start of the programLLAMA_API void llama_backend_init(void);// Call once at the end of the program - currently only used for MPILLAMA_API void llama_backend_free(void);- //optional:+ // Optional: enable numa optimizationsLLAMA_API void llama_numa_init(enum ggml_numa_strategy numa);// Optional: an auto threadpool gets created in ggml if not passed explicitlydiff --git a/src/llama.cpp b/src/llama.cppindex ad8e443882..e97272c889 100644--- a/src/llama.cpp+++ b/src/llama.cpp@@ -401,6 +401,7 @@ static struct llama_model * llama_model_load_from_file_impl(return nullptr;}}+ // TODO: removeggml_time_init();if (!params.vocab_only && ggml_backend_reg_count() == 0) {diff --git a/tests/CMakeLists.txt b/tests/CMakeLists.txtindex 996e7f58f3..01af4ae373 100644--- a/tests/CMakeLists.txt+++ b/tests/CMakeLists.txt@@ -316,6 +316,7 @@ llama_build_and_test(test-backend-sampler.cpp LABEL "model")# Test for state restore with fragmented KV cache# Requires a model, uses same args pattern as test-thread-safety+# TODO: run on all dummy modelsllama_build_and_test(test-state-restore-fragmented.cpp LABEL "model" ARGS -m "${MODEL_DEST}")set_tests_properties(test-state-restore-fragmented PROPERTIES FIXTURES_REQUIRED test-download-model)diff --git a/tests/test-recurrent-state-rollback.cpp b/tests/test-recurrent-state-rollback.cppindex fdac344d89..1944a441cb 100644--- a/tests/test-recurrent-state-rollback.cpp+++ b/tests/test-recurrent-state-rollback.cpp@@ -495,7 +495,7 @@ int main(int argc, char ** argv) {return 1;}- ggml_backend_load_all();+ llama_backend_init();common_init_result_ptr llama_init = common_init_from_params(params);llama_model * model = llama_init->model();diff --git a/tests/test-save-load-state.cpp b/tests/test-save-load-state.cppindex 083ad6a6ca..dee5e17be6 100644--- a/tests/test-save-load-state.cpp+++ b/tests/test-save-load-state.cpp@@ -913,7 +913,7 @@ int main(int argc, char ** argv) {params.n_predict = 16;}- ggml_backend_load_all();+ llama_backend_init();if (!models_dir.empty()) {// run the suite over every dummy model in the directorydiff --git a/tests/test-state-restore-fragmented.cpp b/tests/test-state-restore-fragmented.cppindex d5548afba1..33ce6f2763 100644--- a/tests/test-state-restore-fragmented.cpp+++ b/tests/test-state-restore-fragmented.cpp@@ -30,7 +30,7 @@ int main(int argc, char ** argv) {// init- ggml_backend_load_all();+ llama_backend_init();common_init_result_ptr llama_init = common_init_from_params(params);