<![CDATA[gitblog_09772的博客]]>

<![CDATA[gitblog_09772的博客]]>https://blog.csdn.net/gitblog_09772zh-cnhttps://blog.csdn.net/<![CDATA[GHelper：华硕笔记本轻量硬件控制工具，10MB 单文件替代 500MB 的 Armoury Crate]]>https://blog.csdn.net/gitblog_09772/article/details/143073529https://blog.csdn.net/gitblog_09772/article/details/143073529gitblog_09772Fri, 11 Sep 2026 14:07:12 +0800<![CDATA[GHelper 单个 exe 替代 Armoury Crate：华硕笔记本性能控制全解]]>https://blog.csdn.net/gitblog_09772/article/details/143072733https://blog.csdn.net/gitblog_09772/article/details/143072733gitblog_09772Thu, 10 Sep 2026 09:09:10 +0800<![CDATA[@nuxt/components 与 TypeScript 完美集成：类型安全自动导入的完整教程]]>https://blog.csdn.net/gitblog_09772/article/details/142945890https://blog.csdn.net/gitblog_09772/article/details/142945890gitblog_09772Fri, 17 Jul 2026 07:13:59 +0800<![CDATA[RingProgressBar源码剖析：从零理解Android自定义View绘制原理]]>https://blog.csdn.net/gitblog_09772/article/details/141977496https://blog.csdn.net/gitblog_09772/article/details/141977496gitblog_09772Tue, 23 Jun 2026 08:31:46 +0800<![CDATA[noble-hashes完全指南：如何在JavaScript中安全高效地实现哈希函数、MAC和KDF]]>https://blog.csdn.net/gitblog_09772/article/details/143286788https://blog.csdn.net/gitblog_09772/article/details/143286788gitblog_09772Wed, 17 Jun 2026 08:28:59 +0800<![CDATA[CANN/catlass稀疏数据搬运API]]>https://blog.csdn.net/gitblog_09772/article/details/141946774https://blog.csdn.net/gitblog_09772/article/details/141946774gitblog_09772Fri, 29 May 2026 07:58:28 +0800 [代码位置](https://link.gitcode.com/i/a40a4cc1a6f56629e7fbd382e014e2f6) [TOC] ## 功能说明 `TileCopySparseTla` 在 `copy_l1_to_l0a.hpp` 中定义的偏特化负责 Sparse GEMM 场景下将稀疏化的 A 矩阵 tile 块从 L1（A1 Buffer）搬运到 L0A（A2 Bu]]><![CDATA[CANN/catlass L0C到GM搬运TLA封装]]>https://blog.csdn.net/gitblog_09772/article/details/141947037https://blog.csdn.net/gitblog_09772/article/details/141947037gitblog_09772Fri, 29 May 2026 07:56:01 +0800 [代码位置](https://link.gitcode.com/i/e946bc178085a6a37bb6ddbe99876f54) [TOC] ## 功能说明 `CopyL0CToGmTla` 是 `CopyL0CToGm` 的 TLA 封装，负责将矩阵乘累加结果从 L0C（`CO1`）搬运到 GM，支持类型转换、per-tensor/per-channel 量化/反量化和 ReLU]]><![CDATA[catlass Tile行广播乘法TLA]]>https://blog.csdn.net/gitblog_09772/article/details/141976116https://blog.csdn.net/gitblog_09772/article/details/141976116gitblog_09772Fri, 29 May 2026 07:41:00 +0800 [代码位置](https://link.gitcode.com/i/e2703c5518b988a74415ffc16cbc66cf) [TOC] ## 功能说明 `TileRowBroadcastMulTla` 实现 epilogue 阶段的 TLA 风格广播乘法操作。将 UB 上行向量 (1, n) 广播到 (m, n) 矩阵后与输入逐元素相乘。通过 `ubOut.layout()(]]><![CDATA[CANN ops-blas 开发方案设计]]>https://blog.csdn.net/gitblog_09772/article/details/142133033https://blog.csdn.net/gitblog_09772/article/details/142133033gitblog_09772Fri, 29 May 2026 07:40:35 +0800<![CDATA[CANN/catlass列广播乘法API]]>https://blog.csdn.net/gitblog_09772/article/details/141976251https://blog.csdn.net/gitblog_09772/article/details/141976251gitblog_09772Fri, 29 May 2026 07:40:09 +0800 [代码位置](https://link.gitcode.com/i/c74635a7c64e515228cb21245b2df2d9) [TOC] ## 功能说明 `TileOneBlkColumnBroadcastMul` 实现 epilogue 阶段的列广播乘法操作。将形状 (m, 1) 的列向量在 block 内广播到 (m, n) 后与输入相乘。broadcast 粒度为一个 b]]><![CDATA[GPT2_PMC未来展望：医学AI模型的路线图与社区发展]]>https://blog.csdn.net/gitblog_09772/article/details/141944284https://blog.csdn.net/gitblog_09772/article/details/141944284gitblog_09772Fri, 29 May 2026 07:34:29 +0800<![CDATA[cross-de-zh-roberta-sentence-transformer性能调优：从0.83到0.90的优化之路]]>https://blog.csdn.net/gitblog_09772/article/details/141975842https://blog.csdn.net/gitblog_09772/article/details/141975842gitblog_09772Fri, 29 May 2026 07:28:24 +0800<![CDATA[XLMRoBERTa微调实战：huangjingwang/roberta-ner-multilingual模型训练全流程]]>https://blog.csdn.net/gitblog_09772/article/details/141972957https://blog.csdn.net/gitblog_09772/article/details/141972957gitblog_09772Fri, 29 May 2026 07:27:09 +0800<![CDATA[一文读懂RAFT-stereo：爱芯NPU立体匹配模型的核心优势与应用场景]]>https://blog.csdn.net/gitblog_09772/article/details/141977087https://blog.csdn.net/gitblog_09772/article/details/141977087gitblog_09772Fri, 29 May 2026 07:25:43 +0800<![CDATA[Qwen1.5-7B tokenizer解析：151936词汇表的多语言支持策略]]>https://blog.csdn.net/gitblog_09772/article/details/141977231https://blog.csdn.net/gitblog_09772/article/details/141977231gitblog_09772Fri, 29 May 2026 07:21:46 +0800<![CDATA[E5-base-4k部署完全指南：从本地测试到生产环境的最佳实践]]>https://blog.csdn.net/gitblog_09772/article/details/141977774https://blog.csdn.net/gitblog_09772/article/details/141977774gitblog_09772Fri, 29 May 2026 07:19:46 +0800<![CDATA[bloom-3b-conversational微调指南：如何基于现有模型进行领域适配]]>https://blog.csdn.net/gitblog_09772/article/details/141974320https://blog.csdn.net/gitblog_09772/article/details/141974320gitblog_09772Fri, 29 May 2026 07:15:47 +0800<![CDATA[OpCore-Simplify黑苹果自动化配置工具：5分钟完成OpenCore EFI构建]]>https://blog.csdn.net/gitblog_09772/article/details/143286208https://blog.csdn.net/gitblog_09772/article/details/143286208gitblog_09772Thu, 28 May 2026 17:38:57 +0800<![CDATA[ESP32 Arduino核心库：构建高性能物联网设备的高效解决方案]]>https://blog.csdn.net/gitblog_09772/article/details/143289313https://blog.csdn.net/gitblog_09772/article/details/143289313gitblog_09772Thu, 28 May 2026 17:35:44 +0800<![CDATA[如何彻底掌控你的微信聊天记录？WeChatMsg免费开源解决方案完全指南]]>https://blog.csdn.net/gitblog_09772/article/details/143284070https://blog.csdn.net/gitblog_09772/article/details/143284070gitblog_09772Thu, 28 May 2026 16:29:05 +0800