From 9ae88672f5be8c39ecd91b776c7aebc4dc7fe476 Mon Sep 17 00:00:00 2001 From: coffee Date: Thu, 22 May 2025 18:11:50 +0800 Subject: [PATCH] [2025.05.22] add MoE support --- README.md | 1 + 1 file changed, 1 insertion(+) diff --git a/README.md b/README.md index a25c1d1..b371622 100755 --- a/README.md +++ b/README.md @@ -6,6 +6,7 @@ LLM-Dolly is a custom LLM built from scratch #### 2025-05-22 - 📝 添加支持MoE模型,训练资源占用不稳定,测试模型Experts=8, experts_per_tok=4,GPU显存从60%-94%跳动,单个GPU利用率0%-100%跳动。 +- 当调小batch_size后,显存占用稳定在90%,单个GPU利用率稳定在90%以上,偶现30%的利用率。 #### 2025-05-20 - 📝 添加支持jsonl文件训练.