Lubrication media of rock bit journal bearing is non-Newtonian fluid, and because of structure limitations, lubricant supply is only from the inside track and its rheological properties are relatively complex. In gene...Lubrication media of rock bit journal bearing is non-Newtonian fluid, and because of structure limitations, lubricant supply is only from the inside track and its rheological properties are relatively complex. In general studies, the non- Newtonianism of lubrication media of rock bit journal bearing is neglected to simplify the research process. Based on the universal Reynolds equation of non-Newton media, the mathematical model of lubrication analysis was suitable for the rock bit bearing working conditions was established. Then according to the experimental results of rheological properties of RB-type lubricating grease, the equivalent viscosity of the model was established and this model can be solved by the method used in solving the Reynolds equation of Newton fluid.展开更多
随着大语言模型(large language model,LLM)参数规模的指数级增长,模型部署和推理面临着严峻的内存和计算资源挑战。量化技术作为模型压缩的核心方法,通过降低权重和激活值的数值精度,显著减少了模型的存储需求和计算开销。首先回顾了...随着大语言模型(large language model,LLM)参数规模的指数级增长,模型部署和推理面临着严峻的内存和计算资源挑战。量化技术作为模型压缩的核心方法,通过降低权重和激活值的数值精度,显著减少了模型的存储需求和计算开销。首先回顾了量化技术的发展历程,从经典的Int8/4量化方法到前沿的超低比特量化算法,总结了典型方法的技术特征与性能演进规律,指出传统实数域量化在极低比特条件下存在受限于离散化误差的挑战,难以突破性能上限。为此,进而系统性地梳理了复域量化系列工作。该系列工作提出了基于复数域的量化范式,通过在参数表示中引入幅度与相位2个自由度,显著扩展了模型的表达空间;此外,类比信号处理中通过将时域信号进行傅里叶变换与低通滤波实现稳定表示的经典范式,进一步提出了由实数模型经复域变换与复域量化,达成了无乘法稳定推理的技术路线。实验结果表明,该方案在多个基准数据集上优于现有超低比特量化方法,有效突破了实数域模型的性能天花板,展现出复域量化在高效建模与性能保持方面的潜在价值。总体而言,通过对量化技术演进及复域量化系列研究的系统分析,旨在揭示超低比特量化的发展规律与未来趋势,为高效大模型的理论研究与工程实现提供参考。展开更多
文摘Lubrication media of rock bit journal bearing is non-Newtonian fluid, and because of structure limitations, lubricant supply is only from the inside track and its rheological properties are relatively complex. In general studies, the non- Newtonianism of lubrication media of rock bit journal bearing is neglected to simplify the research process. Based on the universal Reynolds equation of non-Newton media, the mathematical model of lubrication analysis was suitable for the rock bit bearing working conditions was established. Then according to the experimental results of rheological properties of RB-type lubricating grease, the equivalent viscosity of the model was established and this model can be solved by the method used in solving the Reynolds equation of Newton fluid.
文摘随着大语言模型(large language model,LLM)参数规模的指数级增长,模型部署和推理面临着严峻的内存和计算资源挑战。量化技术作为模型压缩的核心方法,通过降低权重和激活值的数值精度,显著减少了模型的存储需求和计算开销。首先回顾了量化技术的发展历程,从经典的Int8/4量化方法到前沿的超低比特量化算法,总结了典型方法的技术特征与性能演进规律,指出传统实数域量化在极低比特条件下存在受限于离散化误差的挑战,难以突破性能上限。为此,进而系统性地梳理了复域量化系列工作。该系列工作提出了基于复数域的量化范式,通过在参数表示中引入幅度与相位2个自由度,显著扩展了模型的表达空间;此外,类比信号处理中通过将时域信号进行傅里叶变换与低通滤波实现稳定表示的经典范式,进一步提出了由实数模型经复域变换与复域量化,达成了无乘法稳定推理的技术路线。实验结果表明,该方案在多个基准数据集上优于现有超低比特量化方法,有效突破了实数域模型的性能天花板,展现出复域量化在高效建模与性能保持方面的潜在价值。总体而言,通过对量化技术演进及复域量化系列研究的系统分析,旨在揭示超低比特量化的发展规律与未来趋势,为高效大模型的理论研究与工程实现提供参考。
