Nearly every recent image synthesis approach, including diffusion,
masked-token prediction, and next-token prediction, uses a Transformer network
architecture. Despite this common backbone, there has been no direct, compute
controlled comparison of how these approaches affect performance and
efficiency. We analyze the scalability of each approach through the lens of
compute budget measured in FLOPs. We find that token prediction methods, led by
next-token prediction, significantly outperform diffusion on prompt following.
On image quality, while next-token prediction initially performs better,
scaling trends suggest it is eventually matched by diffusion. We compare the
inference compute efficiency of each approach and find that next token
prediction is by far the most efficient. Based on our findings we recommend
diffusion for applications targeting image quality and low latency; and
next-token prediction when prompt following or throughput is more important.

最近的图像合成方法主要使用 Transformer 网络架构，本研究通过计算预算的可扩展性对这些方法进行比较，发现基于标记预测的方法在提示后续上明显优于扩散，但在图像质量上，扩散表现与扩大趋势相匹配。论文还比较了每种方法的推理计算效率，发现基于标记预测的方法是最高效的。根据研究发现，建议在追随提示或吞吐量更重要时使用标记预测，而在追求图像质量和低延迟的应用场景中使用扩散。

图像合成中的计算权衡：扩散，掩码令牌和下一个令牌预测

Computational Tradeoffs in Image Synthesis: Diffusion, Masked-Token, and  Next-Token Prediction

This paper presents various automatic detection methods to extract so called
tortured phrases from scientific papers. These tortured phrases, e.g. flag to
clamor instead of signal to noise, are the results of paraphrasing tools used
to escape plagiarism detection. We built a dataset and evaluated several
strategies to flag previously undocumented tortured phrases. The proposed and
tested methods are based on language models and either on embeddings
similarities or on predictions of masked token. We found that an approach using
token prediction and that propagates the scores to the chunk level gives the
best results. With a recall value of .87 and a precision value of .61, it could
retrieve new tortured phrases to be submitted to domain experts for validation.

本文提出了各种自动检测方法来提取科学论文中的所谓被扭曲的短语。这些被扭曲的短语，例如信号噪声代替信号噪声，是为了逃避抄袭检测而使用的释义工具的结果。我们构建了一个数据集，并评估了几种策略来标记以前未记录的被扭曲的短语。所提出和测试的方法基于语言模型，要么基于嵌入相似性，要么基于掩码标记的预测。我们发现，使用标记预测的方法，并将评分传播到块级别可以获得最佳结果。具有 0.87 的召回值和 0.61 的精确值，它可以检索到要提交给领域专家验证的新的被扭曲的短语。