Parallel decoding methods such as Jacobi decoding show promise for more
efficient LLM inference as it breaks the sequential nature of the LLM decoding
process and transforms it into parallelizable computation. However, in
practice, it achieves little speedup compared to traditional autoregressive
(AR) decoding, primarily because Jacobi decoding seldom accurately predicts
more than one token in a single fixed-point iteration step. To address this, we
develop a new approach aimed at realizing fast convergence from any state to
the fixed point on a Jacobi trajectory. This is accomplished by refining the
target LLM to consistently predict the fixed point given any state as input.
Extensive experiments demonstrate the effectiveness of our method, showing
2.4$\times$ to 3.4$\times$ improvements in generation speed while preserving
generation quality across both domain-specific and open-domain benchmarks.

通过改进 Jacobi 解码方法，使其能够在单次迭代中准确预测多个令牌，并在任何状态下快速收敛到固定点，从而实现了在生成速度和生成质量上的显著提升。

CLLMs：一致性大型语言模型

CLLMs: Consistency Large Language Models

In powder diffraction data analysis, phase identification is the process of
determining the crystalline phases in a sample using its characteristic Bragg
peaks. For multiphasic spectra, we must also determine the relative weight
fraction of each phase in the sample. Machine Learning algorithms (e.g.,
Artificial Neural Networks) have been applied to perform such difficult tasks
in powder diffraction analysis, but typically require a significant number of
training samples for acceptable performance. We have developed an approach that
performs well even with a small number of training samples. We apply a
fixed-point iteration algorithm on the labelled training samples to estimate
monophasic spectra. Then, given an unknown sample spectrum, we again use a
fixed-point iteration algorithm to determine the weighted combination of
monophase spectra that best approximates the unknown sample spectrum. These
weights are the desired phase fractions for the sample. We compare our approach
with several traditional Machine Learning algorithms.

本文提出了一种针对粉末衍射数据分析的相同定量分析方法，使用固定点迭代算法进行带标签训练样本的单物相光谱估计，并使用相同算法确定最佳近似未知样本光谱的单物相光谱加权组合，以计算样本相对重量分数，与其他传统机器学习算法进行对比。

基于优化的 PXRD 相分数估计监督学习算法

An Optimization-Based Supervised Learning Algorithm for PXRD Phase  Fraction Estimation

We propose a fast, simple and robust algorithm for computing shortest paths
and distances on Riemannian manifolds learned from data. This amounts to
solving a system of ordinary differential equations (ODEs) subject to boundary
conditions. Here standard solvers perform poorly because they require
well-behaved Jacobians of the ODE, and usually, manifolds learned from data
imply unstable and ill-conditioned Jacobians. Instead, we propose a fixed-point
iteration scheme for solving the ODE that avoids Jacobians. This enhances the
stability of the solver, while reduces the computational cost. In experiments
involving both Riemannian metric learning and deep generative models we
demonstrate significant improvements in speed and stability over both
general-purpose state-of-the-art solvers as well as over specialized solvers.

我们提出了一种快速、简单、健壮的算法，用于计算从数据中学习的黎曼流形上的最短路径和距离。该算法利用常微分方程组（ODE）求解，通过固定点迭代方案来提高求解器的稳定性和降低计算成本。在黎曼度量学习和深度生成模型的实验中，我们证明了该算法相比于最先进的通用求解器和专用求解器在速度和稳定性方面的重大改进。