[2025.06.22] add reward model、PPO、GRPO train code

2025-06-22 22:57:51 +08:00
parent 2a0810012b
commit 4a3c349d33
5 changed files with 423 additions and 14 deletions
@@ -4,11 +4,13 @@ LLM-Dolly is a custom LLM built from scratch

 🗓️ 更新日志

-#### 2025-06-19
- 📝 pretoken corpus and update cold start code.
+#### 2025-06-22
+- 📝 增加奖励模型训练、PPO训练、GRPO训练

 <details>
 <summary>查看更多</summary>
+#### 2025-06-19
+- 📝 pretoken corpus and update cold start code.

 #### 2025-06-05
 - 📝 上传tokenizer的训练权重（1.56字符/token），调整tokenizer的格式和训练方式，和qwen2的tokenizer风格保持一致.