talk-llama : sync llama.cpp

ggml-ci
2025-06-12 20:18:08 +00:00 · 2025-05-13 13:20:19 +03:00
parent a14c89aefa
commit f890560575
25 changed files with 2847 additions and 1125 deletions
--- a/examples/talk-llama/llama-adapter.cpp
+++ b/examples/talk-llama/llama-adapter.cpp
@ -253,6 +253,9 @@ static void llama_adapter_lora_init_impl(llama_model & model, const char * path_
    std::vector<ggml_backend_buffer_type_t> buft_extra;
    {
        auto * cpu_dev = ggml_backend_dev_by_type(GGML_BACKEND_DEVICE_TYPE_CPU);
+        if (!cpu_dev) {
+            throw std::runtime_error(format("%s: no CPU backend found", __func__));
+        }
        auto * cpu_reg = ggml_backend_dev_backend_reg(cpu_dev);

        auto ggml_backend_dev_get_extra_bufts_fn = (ggml_backend_dev_get_extra_bufts_t)
@ -291,6 +294,9 @@ static void llama_adapter_lora_init_impl(llama_model & model, const char * path_
                LLAMA_LOG_WARN("%s: lora for '%s' cannot use buft '%s', fallback to CPU\n", __func__, model_tensor->name, ggml_backend_buft_name(buft));

                auto * cpu_dev = ggml_backend_dev_by_type(GGML_BACKEND_DEVICE_TYPE_CPU);
+                if (!cpu_dev) {
+                    throw std::runtime_error(format("%s: no CPU backend found", __func__));
+                }
                buft = ggml_backend_dev_buffer_type(cpu_dev);

                break;