Small-scale models can achieve competitive summarization results without relying on large language models or human-written references.
