intothephone
intothephone
Thank you very much, man!
> I have just added (in a branch) the version of sesse instead of the original one, and added an example to show the effect. > > You have to...
Got it, thank you very much! Your project is great!
加参数：--lora_target W_pack
Thank you for providing the workaround. But I can't understand your point. Do you mean the entity object supports json format? Which version has this feature? And, even json format...
> > Thank you for providing the workaround. But I can't understand your point. > > Do you mean the entity object supports json format? Which version has this feature?...
一模一样的问题，而且我用的数据还是示例数据，就是那个生成广告的数据集 直接使用原模型秒出回答，但是微调后即使是用数据集里的问题，也是等待200多秒没反应，终于等到有反应了也是回答为空 用户：类型#裤*版型#显瘦*颜色#黑色*风格#简约*裤长#九分裤 ChatGLM-6B： 用户：类型#裤*材质#羊毛*裤长#九分裤*裤口#微喇裤 ChatGLM-6B： 用户：
用的2.7的代码，同样的环境，跑训练没问题，用训练后的模型进行推理也没问题，但转换成推理模型后再预测就出这个CUDNN错误，如之奈何？
Interesting... I've found the root cause: variation To be simple, if variations exist in the pgn, the switching between multiple games will not work. Take following pgn as an example:...
And more interesting, it works fine switching to the game with variations the first time, but after that, you'll never show this game again.