search

zhenhao-huang / CPM-1-Finetune-Text-Generation

Finetune CPM-1 For Text Generation
MIT License
17 stars 3 forks source link

Finetune Issue Tensor 的dim 对应不上 #1

Closed makeme-zgz closed 3 years ago

makeme-zgz commented 3 years ago

File "/home/ubuntu/cpm/mpu/transformer.py", line 132, in forward attention_scores = torch.mul(attention_scores, ltor_mask) - \ RuntimeError: The size of tensor a (36) must match the size of tensor b (648) at non-singleton dimension 3

zhenhao-huang commented 3 years ago

您可以提供更详细的错误信息吗

makeme-zgz commented 3 years ago

替换了deepspeed的版本之后就没有这个问题了。 0.3.15 的deepspeed是可以的