1. test huggingface manually from IPython;
1.1. seems to work;
1.2 put some long text from documentation,
it has provided some summary;
1.3. runs in about 40-60 seconds on 3-4 cores CPU;
1.4. almost made reuse local cache,
for some reason huggingface still can
download LFS weights from the public repo;
2. partially wrapped into worker
which is to be run as a separate service in docker-compose;