From 9e56e6c50dc01026f61db1e362f460efeda71d2e Mon Sep 17 00:00:00 2001 From: He Yongzhe Date: Mon, 30 Oct 2023 19:42:34 +0800 Subject: [PATCH] Update [WeeklyReport]2023.10.10~2023.10.24.md add comment --- .../12_Corle-hyz/[WeeklyReport]2023.10.10~2023.10.24.md | 1 + 1 file changed, 1 insertion(+) diff --git a/WeeklyReports/12_Corle-hyz/[WeeklyReport]2023.10.10~2023.10.24.md b/WeeklyReports/12_Corle-hyz/[WeeklyReport]2023.10.10~2023.10.24.md index 1116ef3e..f9ca58b2 100644 --- a/WeeklyReports/12_Corle-hyz/[WeeklyReport]2023.10.10~2023.10.24.md +++ b/WeeklyReports/12_Corle-hyz/[WeeklyReport]2023.10.10~2023.10.24.md @@ -42,3 +42,4 @@ Github ID：[Corle-hyz](https://github.com/Corle-hyz) 1. 明确Llama相较于Transformer修改的地方，仿照[《Reducing Activation Recomputation in Large Transformer Models》](https://arxiv.org/abs/2205.05198)的方式，从理论推导给出Llama的激活内存模型。 ### 导师点评 +咏哲同学基础扎实，能够快速适应全自动并行任务，本周在显存建模工作中提出不错的想法，需要进一步结合已有的工业实践进一步完善建模工作，达到可大范围推广的目标。期待咏哲的下一步的工作成果。