Function load_gpu_split_from_split_file

llama.cpp:3082–3086

};

static bool load_gpu_split_from_split_file(llama_model & model, std::string split_path, size_t vram_budget) {
    llama_gpu_split_loader loader(split_path, true);
    return loader.check_vram_allocable(vram_budget) 
        && loader.load_gpu_idx_for_model(&model) == 0;
}

static bool llm_load_gpu_split_with_budget(llama_model_loader & ml, llama_model & model, size_t vram_allocatable_bytes, bool no_cache) {
    std::string cached_split_path = ml.file.fname + ".generated.gpuidx";