Method pos_next

tools/server/server-common.cpp:246–287

}

llama_pos server_tokens::pos_next(int64_t n_tokens) const {
    if (!has_mtmd) {
        if (n_tokens < 0) {
            return tokens.size();
        }

        return n_tokens;
    }

    if (n_tokens < 0) {
        llama_pos res = tokens.size();

        for (auto it = map_idx_to_media.begin(); it != map_idx_to_media.end(); ++it) {
            const auto & chunk = it->second;
            res += mtmd_input_chunk_get_n_pos(chunk.get()) - mtmd_input_chunk_get_n_tokens(chunk.get());
        }

        return res;
    }

    int64_t idx = 0;
    llama_pos pos = 0;

    GGML_ASSERT(n_tokens <= (int64_t)tokens.size());

    while (idx < n_tokens) {
        const auto media_it = map_idx_to_media.find(idx);
        if (media_it != map_idx_to_media.end()) {
            const auto & chunk = media_it->second;
            const llama_pos n_pos = mtmd_input_chunk_get_n_pos(chunk.get());
            const size_t n_tok = mtmd_input_chunk_get_n_tokens(chunk.get());

            pos += n_pos;
            idx += n_tok;
        } else {
            pos++;
            idx++;
        }
    }

    return pos;
}

size_t server_tokens::size_up_to_pos(llama_pos max_pos) const {
    if (!has_mtmd) {