Onepager

DeepSeek, GPU Platforms, LoRA, and Positional Embedding

This page keeps the onepager reading experience while the body content is maintained in four markdown sources. The page unfolds in a fixed order: DeepSeekMoE and V3 parallel training, a data center GPU platform comparison across H100, GB200, and GB300, LoRA and MLA, and positional embedding.