<?xml version="1.0" encoding="utf-8" standalone="yes"?><urlset xmlns="http://www.sitemaps.org/schemas/sitemap/0.9" xmlns:xhtml="http://www.w3.org/1999/xhtml"><url><loc>https://niyunsheng.github.io/tech/</loc><lastmod>2026-08-18T10:00:00+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/categories/</loc><lastmod>2026-08-18T10:00:00+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/tags/deepseek/</loc><lastmod>2026-08-18T10:00:00+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/tags/dsa/</loc><lastmod>2026-08-18T10:00:00+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/deepseek-mla-dsa-hca-csa/</loc><lastmod>2026-08-18T10:00:00+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/tags/kv-cache/</loc><lastmod>2026-08-18T10:00:00+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/tags/mla/</loc><lastmod>2026-08-18T10:00:00+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/categories/model-architecture/</loc><lastmod>2026-08-18T10:00:00+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/series/open-model-architectures/</loc><lastmod>2026-08-18T10:00:00+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/series/</loc><lastmod>2026-08-18T10:00:00+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/tags/sparse-attention/</loc><lastmod>2026-08-18T10:00:00+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/tags/</loc><lastmod>2026-08-18T10:00:00+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/</loc><lastmod>2026-08-18T10:00:00+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/tags/dit/</loc><lastmod>2026-08-06T12:00:00+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/rope-and-timestep-embedding/</loc><lastmod>2026-08-06T12:00:00+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/tags/rope/</loc><lastmod>2026-08-06T12:00:00+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/tags/timestep-embedding/</loc><lastmod>2026-08-06T12:00:00+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/categories/video-generation/</loc><lastmod>2026-08-06T12:00:00+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/series/video-generation-theory/</loc><lastmod>2026-08-06T12:00:00+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/absorbed-mla-vs-naive-mla/</loc><lastmod>2026-08-02T09:30:00+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/tags/flops/</loc><lastmod>2026-08-02T09:30:00+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/tags/roofline/</loc><lastmod>2026-08-02T09:30:00+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/categories/system-optimization/</loc><lastmod>2026-08-02T09:30:00+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/tags/bank-conflict/</loc><lastmod>2026-03-22T08:00:00+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/tags/cuda/</loc><lastmod>2026-03-22T08:00:00+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/tags/gemm/</loc><lastmod>2026-03-22T08:00:00+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/series/gemm-optimization/</loc><lastmod>2026-03-22T08:00:00+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/categories/kernels/</loc><lastmod>2026-03-22T08:00:00+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/cuda-gemm-bank-conflict-swizzling/</loc><lastmod>2026-03-22T08:00:00+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/tags/shared-memory/</loc><lastmod>2026-03-22T08:00:00+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/tags/swizzling/</loc><lastmod>2026-03-22T08:00:00+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/tags/context-parallelism/</loc><lastmod>2026-03-21T15:41:55+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/categories/distributed-system/</loc><lastmod>2026-03-21T15:41:55+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/loss-reduction-dp-cp/</loc><lastmod>2026-03-21T15:41:55+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/tags/parallelism/</loc><lastmod>2026-03-21T15:41:55+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/global-grad-norm-hybrid-parallel/</loc><lastmod>2026-03-21T07:02:02+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/tags/fsdp/</loc><lastmod>2026-03-21T07:02:02+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/tags/gradient-clipping/</loc><lastmod>2026-03-21T07:02:02+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/flash-attention-forward-backward-triton/</loc><lastmod>2026-03-15T22:00:00+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/tags/flash-attention/</loc><lastmod>2026-03-15T22:00:00+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/tags/triton/</loc><lastmod>2026-03-15T22:00:00+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/tags/flow-matching/</loc><lastmod>2026-02-25T22:00:00+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/video-training-tricks/</loc><lastmod>2026-02-25T22:00:00+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/tags/training-stability/</loc><lastmod>2026-02-25T22:00:00+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/triton-gemm-tiling-tma/</loc><lastmod>2026-02-11T00:39:00+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/tags/tiling/</loc><lastmod>2026-02-11T00:39:00+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/tags/tma/</loc><lastmod>2026-02-11T00:39:00+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/tags/h200/</loc><lastmod>2026-02-09T23:20:00+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/tags/mfu/</loc><lastmod>2026-02-09T23:20:00+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/series/performance-modeling/</loc><lastmod>2026-02-09T23:20:00+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/tags/recomputation/</loc><lastmod>2026-02-09T23:20:00+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/roofline-llm-h200/</loc><lastmod>2026-02-09T23:20:00+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/tags/classifier-free-guidance/</loc><lastmod>2026-02-03T23:30:00+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/tags/ddpm/</loc><lastmod>2026-02-03T23:30:00+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/ddpm-to-flow-matching/</loc><lastmod>2026-02-03T23:30:00+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/adaln-zero-dit-sd3-hunyuan/</loc><lastmod>2026-02-02T23:30:00+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/llm-flops-mfu-hfu/</loc><lastmod>2026-01-31T09:30:00+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/series/3d-sparse-attention/</loc><lastmod>2026-01-26T23:30:00+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/tags/visualization/</loc><lastmod>2026-01-26T23:30:00+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/3d-sliding-window-attention-viz/</loc><lastmod>2026-01-26T23:30:00+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/gpu-network-constants/</loc><lastmod>2026-01-25T23:21:35+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/tags/gpu-specs/</loc><lastmod>2026-01-25T23:21:35+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/categories/hardware/</loc><lastmod>2026-01-25T23:21:35+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/tags/interconnect/</loc><lastmod>2026-01-25T23:21:35+08:00</lastmod></url><url><loc>https://niyunsheng.github.io/archives/</loc></url><url><loc>https://niyunsheng.github.io/explore/</loc></url><url><loc>https://niyunsheng.github.io/guestbook/</loc></url><url><loc>https://niyunsheng.github.io/search/</loc></url></urlset>