<?xml version="1.0" encoding="utf-8" ?><rss version="2.0"><channel><title><![CDATA[BerryRoad的博客]]></title><description><![CDATA[]]></description><link>https://blog.youkuaiyun.com/BerryRoad</link><language>zh-cn</language><generator>https://blog.youkuaiyun.com/</generator><copyright><![CDATA[Copyright &copy; BerryRoad]]></copyright><item><title><![CDATA[Datawhale Quant For Beginners 学习笔记3]]></title><link>https://blog.youkuaiyun.com/BerryRoad/article/details/163007101</link><guid>https://blog.youkuaiyun.com/BerryRoad/article/details/163007101</guid><author>BerryRoad</author><pubDate>Sun, 19 Jul 2026 02:56:18 +0800</pubDate><description><![CDATA[今天学了哪个task完成了哪些课程要求一段运行结果或学习记录一个还没完全懂的问题价格与移动平均线任务2：MA10/MA30 金叉死叉标注图的标题不要带。]]></description><category></category></item><item><title><![CDATA[Datawhale Quant For Beginners 学习笔记2]]></title><link>https://blog.youkuaiyun.com/BerryRoad/article/details/162946529</link><guid>https://blog.youkuaiyun.com/BerryRoad/article/details/162946529</guid><author>BerryRoad</author><pubDate>Thu, 16 Jul 2026 23:11:28 +0800</pubDate><description><![CDATA[以前面苹果、英伟达、特斯拉的收益率数据为例：特斯拉标准差（σ = 2.83%）明显高于苹果（σ = 1.52%），它的 Histogram 尾巴也更长。：美国独立日休市，因7月4日是周六，休市提前至7月3日（周五）。在本实验设定下（1y 日线），波动最大的是：哔哩哔哩（美股）查询日期：2026-07-14 (UTC+8)查询日期：2026-07-14 (UTC+8)查询日期：2026-07-14 (UTC+8)查询日期：2026-07-14 (UTC+8)查询日期：2026-07-14 (UTC+8)]]></description><category></category></item><item><title><![CDATA[Datawhale Quant For Beginners 学习笔记 1]]></title><link>https://blog.youkuaiyun.com/BerryRoad/article/details/162850522</link><guid>https://blog.youkuaiyun.com/BerryRoad/article/details/162850522</guid><author>BerryRoad</author><pubDate>Mon, 13 Jul 2026 22:05:43 +0800</pubDate><description><![CDATA[数据区间：2026-01-12 ~ 2026-07-10（约 6 个月）]]></description><category></category></item><item><title><![CDATA[helloROCm-Day2-体验模型微调]]></title><link>https://blog.youkuaiyun.com/BerryRoad/article/details/162078553</link><guid>https://blog.youkuaiyun.com/BerryRoad/article/details/162078553</guid><author>BerryRoad</author><pubDate>Wed, 17 Jun 2026 23:59:08 +0800</pubDate><description><![CDATA[安装依赖时vllm、triton、torch下载速度比较慢，打开新的终端窗口从魔搭同步下载模型。下载后替换原本下载模型的步骤，直接给 LOCAL_MODEL_DIR 赋值。可以打开新的终端手动安装，然后直接从下面开始执行.使用Notebook执行显示的信息有缩减。使用命令行执行有更丰富的输出内容。]]></description><category></category></item><item><title><![CDATA[helloROCm-Day1-Gemma4-部署笔记]]></title><link>https://blog.youkuaiyun.com/BerryRoad/article/details/161962661</link><guid>https://blog.youkuaiyun.com/BerryRoad/article/details/161962661</guid><author>BerryRoad</author><pubDate>Sat, 13 Jun 2026 23:50:23 +0800</pubDate><description><![CDATA[torch.cudaTRITON_ATTN = 用 OpenAI Triton 语言写的注意力计算内核↓优点：跨平台（NVIDIA/AMD）、可灵活适配特殊模型结构↓当前场景：Gemma 4 有混合头维度 + AMD ROCm 显卡↓结果：vLLM 自动选择 TRITON_ATTN 作为注意力后端。]]></description><category></category></item></channel></rss>