Method state_read

subprojects/llama.cpp/src/llama-kv-cache.cpp:1697–1733

}

void llama_kv_cache::state_read(llama_io_read_i & io, llama_seq_id seq_id, llama_state_seq_flags flags) {
    GGML_UNUSED(flags);

    GGML_ASSERT(seq_id == -1 || (seq_id >= 0 && (size_t) seq_id < seq_to_stream.size()));

    uint32_t n_stream_cur;
    io.read_to(&n_stream_cur, sizeof(n_stream_cur));
    if (n_stream_cur != n_stream) {
        throw std::runtime_error("n_stream mismatch");
    }

    for (uint32_t s = 0; s < n_stream; ++s) {
        uint32_t cell_count;
        io.read_to(&cell_count, sizeof(cell_count));

        if (cell_count == 0) {
            continue;
        }

        const uint32_t strm = seq_id == -1 ? s : seq_to_stream[seq_id];

        slot_info sinfo;

        bool res = true;
        res = res && state_read_meta(io, strm, cell_count, sinfo, seq_id);
        res = res && state_read_data(io, strm, cell_count, sinfo);

        if (!res) {
            if (seq_id == -1) {
                clear(true);
            } else {
                seq_rm(seq_id, -1, -1);
            }
            throw std::runtime_error("failed to restore kv cache");
        }
    }
}

void llama_kv_cache::state_write_meta(llama_io_write_i & io, const cell_ranges_t & cr, llama_seq_id seq_id) const {
    const auto & cells = v_cells[cr.strm];