在神经网络研究的前沿，我们正面临着模型精度与运行效率之间的权衡挑战。尽管架构优化、层融合和模型编译等技术已取得显著进展，但这些方法往往不足以同时满足边缘设备部署所需的模型尺寸和精度要求。 研究人员通常采用三种主要策略来实现模型压缩 ...

DeepSeek在发布其V3.1大语言模型时宣布该模型采用了"UE8M0 FP8 scale data format"进行训练，这一技术细节引发了业界对于新兴量化格式的广泛关注。UE8M0作为FP8格式家族中的一个特殊变体，我们今天来看看这个UE8M0到底是什么。 数值表示格式 我们先看看一般的量化表示 ...

网上关于算力的文章，如果提到某个芯片或某个智算中心的算力，都会写： 在FP32精度下，英伟达H100的算力大约为 0.9 PFlops ...

The feature image you see above was generated by an AI text-to-image rendering model called Stable Diffusion. Stable Diffusion typically runs in the cloud via a web browser, and is driven by data ...

The general definition of quantization states that it is the process of mapping continuous infinite values to a smaller set of discrete finite values. In this blog, we will talk about quantization in ...

INT8 provides better performance with comparable precision than floating point for AI inference. But when INT8 is unable to meet the desired performance with limited resources, INT4 optimization is ...