Reproduce “Generation with Dynamic Vocabulary” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yanting Liu, Tao Ji, Changzhi Sun, Yuanbin Wu, Xiaoling Wang
The method extends a language model’s fixed vocabulary with retrieved phrases it can select and consume as single steps, without changing backbone parameters. Phrases are encoded for both input and output. On WikiText-103, it reports MAUVE 25.69 and decoding latency 0.99 seconds, versus 20.47 and 1.10 for GPT-2; latency covers decoding, not end-to-end serving.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.