This paper introduces Block Data Representations (BDR), a framework for exploring and evaluating a wide spectrum of narrow-precision formats for deep learning. It enables comparison of popular quantization standards, and through BDR, new formats based on shared microexponents (MX) are identified, which outperform other state-of-the-art quantization approaches, including narrow-precision floating-point and block floating-point. MX utilizes multiple levels of quantization scaling with ultra-fine scaling factors based on shared microexponents in the hardware. The effectiveness of MX is demonstrated on real-world models including large-scale generative pretraining and inferencing, and production-scale recommendation systems.
Mesoporous silica and titania supraparticles with controllable pore size, particle size, and macroscopic morphology were readily synthesized by a novel synthetic pathway using meniscus templating on a superhydrophobic surface, which is much simpler than well-known emulsion systems. Moreover, we first report that despite the very large radius of droplet curvature on a millimeter scale, supraparticles kept the round cap morphology due to addition of sucrose as a shape preserver as well as a pore-forming agent. In addition, mesoporous silica and titania supraparticles provided good adsorption performance for Acid Blue 25 and Cr(VI), and were easily separated from the solution by using a scoop net after adsorption tests.