Philrain
> In the example of using the dowhy toolkit to estimate the effect of the loyalty reward program, I used my own data, and other program commands were the same,...
I also encountered this problem, have you solved it? Can you tell me how to solve it?
> I can conduct experiments directly using dowhy's tools, but I need to use EconML's machine learning library for this experiment, so I need to use EconML's fit() function, so...
> Hi, could you please include the code that you were running when you saw this issue, the full stack trace, and the output of `pip list`? Thanks! aiohttp 3.7.4.post0...
> @xiongma 碰到了同样的问题，第一个epoch=2.3，然后loss=0，加载权重后发现很多层都变成了nan。我感觉是FP16在全量微调下不稳定可能造成了梯度爆炸。我在模型初始化后加了一步model.float(), 把模型参数调整成了FP32，再进行微调loss就正常了。但确实是既慢。。。又大。。。。 我调整成fp32之后又说我输入也要保持一致，但是input_ids, attention_mask也没办法是float32呀
请问解决了吗？
我也是星火不行，渠道里测试都不行
我也是星火不行，渠道里测试都不行
请问您现在知道了么,我也是刚学,想请教一下
> 你的词典文件没有改名 需要改为vocab.txt 已经改的就是这个了呀