Update README.md
Browse files
README.md
CHANGED
@@ -36,10 +36,15 @@ In the term "deepsex," the Chinese character "sex" (性) refers to:
|
|
36 |
|
37 |
Please note this "sex" should not be interpreted as sexual relations, but rather as the multidimensional Chinese philosophical concept of "Xing" (性).
|
38 |
|
39 |
-
##
|
40 |
|
41 |
-
|
42 |
-
|
|
|
|
|
|
|
|
|
|
|
43 |
|
44 |
## 鸣谢
|
45 |
- **上海左北科技提供算法与算力和语料**[企业网址](https://leftnorth.com/)
|
|
|
36 |
|
37 |
Please note this "sex" should not be interpreted as sexual relations, but rather as the multidimensional Chinese philosophical concept of "Xing" (性).
|
38 |
|
39 |
+
## 更新记录
|
40 |
|
41 |
+
- 0222更新-(进度23%,双版本发布,增加普通版)
|
42 |
+
|
43 |
+
- 为了解决部分平台不兼容模型的问题,去掉思维链训练了一个普通版本,为NoCot版,同样采用MGRPO策略训练,但可能效果不及Cot版,也可能上下文连贯性好于Cot版。
|
44 |
+
- Cot版本训练进度为23%,已训练完成20万轮对话,剩余约70万轮,loss下降至1.1,即将接近7b模型极限。
|
45 |
+
|
46 |
+
- 0218更新-(进度15%,动态思维链)
|
47 |
+
- Tifa-DeepsexV2-7b-MGRPO-0218,大幅减少发散、提高推理能力、惩罚多个过拟合词,自动选择是否进行深度思考(测试)
|
48 |
|
49 |
## 鸣谢
|
50 |
- **上海左北科技提供算法与算力和语料**[企业网址](https://leftnorth.com/)
|