GPT2-Chinese
GPT2-Chinese copied to clipboard
多GPU训练with attention_mask transformers/modeling_gpt2.py line 403 StopIteration
版本:
python 3.6.0
torch 1.6.0+cu101
在多GPU without attention_mask的时候一切正常, 但是加上attention_mask后在forward阶段报错如下:
File "$PATH/envs/gpt_venv/lib/python3.6/site-packages/transformers/modeling_gpt2.py", line 403, in forward
attention_mask = attention_mask.to(dtype=next(self.parameters()).dtype) # fp16 compatibility
StopIteration
请问有没有人遇到过相似的错误呢?
debug了一下尝试了不同的torch版本,torch==1.4, 1.5均会报错Segmentation fault (core dumped)