b4566
llama : minor fixes for up llama load model speed (#11448) * impl::load change map bpe_ranks to onordered map for reduce time of impl::load on 30% * llama_model_loader::init_mapping - replace new llama_mmap to std::make_unique<llama_mmap> for clean code & reduce (/2) time of running init_mappings * Update src/llama-vocab.cpp --------- Co-authored-by: lexasub <empty@empty.ru> Co-authored-by: Diego Devesa <slarengh@gmail.com>