From e4da44fcdc41a64246d752f83634b90c2aea824e Mon Sep 17 00:00:00 2001 From: p04896573 Date: Mon, 19 Aug 2024 11:07:45 +0800 Subject: [PATCH] Update README.md --- README.md | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-) diff --git a/README.md b/README.md index dced065..224e19b 100644 --- a/README.md +++ b/README.md @@ -9,8 +9,7 @@ ## 2024.08.19 NOTICE - 由于新架构中多数据集验证发现2B模型进行lora训练效果不及全参数微调 - 2B模型采用全参数微调训练,我们在[QUICK START](https://www.osredm.com/jiuyuan/CPM-9G-8B/tree/FM_9G/quick_start_clean/readmes/quick_start.md) 中更新了更多关于微调训练的信息 -- 8B模型LORA微调在master分支进行 - +- 8B模型LORA微调在master分支进行训练 ## 版本更新内容 具体的迭代信息如下: - 训练:升级了训练代码,提升GPU利用率和并行化,并且2B模型能兼容transformers中的tokenizer(LlamaTokenizerFast)