yaoyaoyao
yaoyaoyao
It seems that Baicizhan has closed the API without any annoucement. : (
只需要在OpenLink Virtuoso加载你自己的数据,然后设置对应的SPARQLPATH就行,不过当前的prompt都是针对freebase调的,如果使用自己的知识图谱可能需要对prompt进行一些调整。
I face the same issue, the problem is in torch.distributed.all_gather. But I fail to solve it even set NCCL_P2P_DISABLE=1.
I have found the issue, the sizes of tensors on different GPU are different in my codes, so the program stucks without any error or warning.