Chong Ruan

单机 A100 是几张卡？打开 CUDA_LAUNCH_BLOCKING=1 试试呢，报错在哪里？

最近很多人问这个问题，我们正在研究放一个正式的 tokenizer.model 出来。着急的话可以先用 TheBloke 转好的模型，但他的分词结果和我们应该是有一些差异的，所以效果应该会比用官方 python 代码跑出来的结果差一些。

We don't run it on Mac internally, so nobody is doing this. But I guess this should be easy and just need a little bit coding. We would appreciate it...

Glad to hear that and waiting for your PR.

I don't see why. I guess this error is irrelavant. It might come from your other local repositories. Is there any detailed error information?

好多年了，早忘了= = 如果要排查的话，建议先把数据集弄小一点，比如说就一个正样本一个负样本开始试，然后一步一步扩大数据集规模。