Function llama_adapter_lora_init_impl

smallthinker/src/llama-adapter.cpp:148–369

}

static void llama_adapter_lora_init_impl(llama_model & model, const char * path_lora, llama_adapter_lora & adapter) {
    LLAMA_LOG_INFO("%s: loading lora adapter from '%s' ...\n", __func__, path_lora);

    ggml_context * ctx_init;
    gguf_init_params meta_gguf_params = {
        /* .no_alloc = */ true,
        /* .ctx      = */ &ctx_init,
    };

    gguf_context_ptr ctx_gguf { gguf_init_from_file(path_lora, meta_gguf_params) };
    if (!ctx_gguf) {
        throw std::runtime_error("failed to load lora adapter file from " + std::string(path_lora));
    }

    ggml_context_ptr ctx { ctx_init };

    // check metadata
    {
        auto get_kv_str = [&](const std::string & key) -> std::string {
            int id = gguf_find_key(ctx_gguf.get(), key.c_str());
            return id < 0 ? "" : std::string(gguf_get_val_str(ctx_gguf.get(), id));
        };
        auto get_kv_f32 = [&](const std::string & key) -> float {
            int id = gguf_find_key(ctx_gguf.get(), key.c_str());
            return id < 0 ? 0.0f : gguf_get_val_f32(ctx_gguf.get(), id);
        };
        LLM_KV llm_kv = LLM_KV(LLM_ARCH_UNKNOWN);

        auto general_type = get_kv_str(llm_kv(LLM_KV_GENERAL_TYPE));
        if (general_type != "adapter") {
            throw std::runtime_error("expect general.type to be 'adapter', but got: " + general_type);
        }

        auto general_arch_str = get_kv_str(llm_kv(LLM_KV_GENERAL_ARCHITECTURE));
        auto general_arch = llm_arch_from_string(general_arch_str);
        if (general_arch != model.arch) {
            throw std::runtime_error("model arch and LoRA arch mismatch");
        }

        auto adapter_type = get_kv_str(llm_kv(LLM_KV_ADAPTER_TYPE));
        if (adapter_type != "lora") {
            throw std::runtime_error("expect adapter.type to be 'lora', but got: " + adapter_type);
        }

        adapter.alpha = get_kv_f32(llm_kv(LLM_KV_ADAPTER_LORA_ALPHA));
    }

    int n_tensors = gguf_get_n_tensors(ctx_gguf.get());

    // contexts for each buffer type
    std::map<ggml_backend_buffer_type_t, ggml_context *> ctx_map;
    auto ctx_for_buft = [&](ggml_backend_buffer_type_t buft) -> ggml_context * {
        auto it = ctx_map.find(buft);
        if (it == ctx_map.end()) {
            // add a new context
            ggml_init_params params = {
                /*.mem_size   =*/ n_tensors*ggml_tensor_overhead(),
                /*.mem_buffer =*/ NULL,