Wenaka

Results 3 issues of Wenaka

**问题描述** 在使用deepseek r1模型进行对话时,程序会把推理模型回复的字段当成标题,导致标题显示异常无法轻易获取对话主旨 **解决思路** 在设置里提供指定用于生成对话标题的模型选项 **附加上下文** ![Image](https://github.com/user-attachments/assets/8635f4cf-62da-4f5d-9dc5-1428bcc2f2c2)

When training with `cache_text_embeddings: true` and sampling enabled (which is the default), the training fails with: ``` AttributeError: 'FakeTextEncoder' object has no attribute 'encoder' ``` This occurs during the initial...