PARAGEN : A Parallel Generation Toolkit
This toolkit addresses the need for efficient experimentation and deployment in parallel generation for NLP researchers and industry practitioners, though it is incremental as it builds on existing frameworks.
The authors introduced PARAGEN, a PyTorch-based NLP toolkit designed to facilitate development in parallel generation, offering customizable plugins and features like unlimited data loading and automatic model selection to support research and industry applications at ByteDance.
PARAGEN is a PyTorch-based NLP toolkit for further development on parallel generation. PARAGEN provides thirteen types of customizable plugins, helping users to experiment quickly with novel ideas across model architectures, optimization, and learning strategies. We implement various features, such as unlimited data loading and automatic model selection, to enhance its industrial usage. ParaGen is now deployed to support various research and industry applications at ByteDance. PARAGEN is available at https://github.com/bytedance/ParaGen.