Investigating Efficiently Extending Transformers for Long Input Summarization
This addresses the problem of long input summarization for NLP practitioners by providing an efficient extension to existing models, though it is incremental as it builds on prior work like PEGASUS.
The paper tackles the challenge of adapting pretrained Transformers for long input summarization by investigating architectural changes and pretraining paradigms, resulting in PEGASUS-X, which handles up to 16K tokens and achieves strong performance comparable to larger models with few additional parameters and no model parallelism.
While large pretrained Transformer models have proven highly capable at tackling natural language tasks, handling long sequence inputs continues to be a significant challenge. One such task is long input summarization, where inputs are longer than the maximum input context of most pretrained models. Through an extensive set of experiments, we investigate what model architectural changes and pretraining paradigms can most efficiently adapt a pretrained Transformer for long input summarization. We find that a staggered, block-local Transformer with global encoder tokens strikes a good balance of performance and efficiency, and that an additional pretraining phase on long sequences meaningfully improves downstream summarization performance. Based on our findings, we introduce PEGASUS-X, an extension of the PEGASUS model with additional long input pretraining to handle inputs of up to 16K tokens. PEGASUS-X achieves strong performance on long input summarization tasks comparable with much larger models while adding few additional parameters and not requiring model parallelism to train.