Replies: 21 comments 15 replies
-
这很可能是模型生成了工具调用或代码执行等包含 metadata 的对话内容。目前 tokenizer 中对 也许可以考虑运行 综合 Demo ? |
Beta Was this translation helpful? Give feedback.
-
综合demo不退出了,但是仍然没法正常对话,输出是混乱的,我往前翻了issue,有一个相同的问题 #71 |
Beta Was this translation helpful? Give feedback.
-
chatglm3-6b-base也有同样的问题,发生同样的位置。调整输入后故障可能消失,而且有时model.chat返回的0号元素会变成一个json对象而不是字符串。 |
Beta Was this translation helpful? Give feedback.
-
对于这个出现乱码的情况,想确认一下您用的是 |
Beta Was this translation helpful? Give feedback.
-
确实是chatglm3-6b,还测试了chatglm3-6b-32k,一样的现象,我先试一下上面 @vcvcvnvcvcvn 提到的方法 |
Beta Was this translation helpful? Give feedback.
-
@0000sir 嗯嗯,我的做法只能解决"ValueError: not enough values to unpack"的问题,关于乱码是什么情况就不知道了 |
Beta Was this translation helpful? Give feedback.
-
使用后倒是解决"ValueError: not enough values to unpack"的问题了,但有时候会出现TypeError: unhashable type: 'slice' |
Beta Was this translation helpful? Give feedback.
-
确实,乱码问题仍然存在,我试试换个cuda版本看看 |
Beta Was this translation helpful? Give feedback.
-
换 nvidia/cuda:12.1.0-runtime-ubuntu22.04 仍然是乱码 |
Beta Was this translation helpful? Give feedback.
-
检查一下lora方式对不对?glm3的微调格式与其他不同 |
Beta Was this translation helpful? Give feedback.
-
该问题是一个bad case,目前已经移到Discussion,我们将继续收集这一类问题并解决 |
Beta Was this translation helpful? Give feedback.
-
现在有人尝试使用fschat等实现多卡推理成功的么 |
Beta Was this translation helpful? Give feedback.
-
我也遇到这个问题。。。我用2张p4跑,无论是官方的脚本还是用fastchat,都输出无意义的字符,跑的chatglm3-6b |
Beta Was this translation helpful? Give feedback.
-
chatglm3模型推理,如何保证同样的问题输出是一样的?固定随机种子也没有效果 |
Beta Was this translation helpful? Give feedback.
-
Beta Was this translation helpful? Give feedback.
-
请问这个 load_model_on_gpus 函数在哪里怎么找不到? |
Beta Was this translation helpful? Give feedback.
-
3060 12G双显卡,在修改代码后,运行cli_demo.py对话时提示 "ValueError: not enough values to unpack"
以下是错误内容:
测试在Tesla P100上双卡运行没有问题
3060驱动版本 545.23.06
P100区动版本 530.30.02
报错的第48行是这个循环的开始,应该是model.stream_chat调用失败
请问这可能是什么原因,如何修复
Beta Was this translation helpful? Give feedback.
All reactions