S O'Brien, M Lewis - arXiv preprint arXiv:2309.09117, 2023 - arxiv.org
We demonstrate that Contrastive Decoding--a simple, computationally light, and training- free text generation method proposed by Li et al 2022--achieves large out-of-the-box …
The dominant paradigm for neural text generation is left-to-right decoding from autoregressive language models. Constrained or controllable generation under complex …
Given an input sequence (or prefix), modern language models often assign high probabilities to output sequences that are repetitive, incoherent, or irrelevant to the prefix; as …
RY Pang, H He - arXiv preprint arXiv:2009.07839, 2020 - arxiv.org
Current approaches to text generation largely rely on autoregressive models and maximum likelihood estimation. This paradigm leads to (i) diverse but low-quality samples due to …
Large pre-trained language models have repeatedly shown their ability to produce fluent text. Yet even when starting from a prompt, generation can continue in many plausible …
Y Su, N Collier - arXiv preprint arXiv:2210.14140, 2022 - arxiv.org
Generating text with autoregressive language models (LMs) is of great importance to many natural language processing (NLP) applications. Previous solutions for this task often …
In open-ended natural-language generation, existing text decoding methods typically struggle to produce text which is both diverse and high-quality. Greedy and beam search are …
Pretrained language models (LMs) are susceptible to generate text with nonfactual information. In this work, we measure and improve the factual accuracy of large-scale LMs …
Despite recent advances in natural language generation, it remains challenging to control attributes of generated text. We propose DExperts: Decoding-time Experts, a decoding-time …