GitLab

- Convert use of deprecated AutoModelWithLMHead to AutoModelForSeq2SeqLM
- Add newly required `truncation=True` to `tokenizer.encode` with `max_length`

This silences all warnings.