Runtime error in backward

[ridiculouz]

你好，在尝试使用这份代码时，我遇到了如下的bug： Traceback (most recent call last):
File "run.py", line 64, in trainer.train() File "/home/anaconda3/envs/pytorch/lib/python3.8/site-packages/fastNLP/core/trainer.py", line 622, in train raise e File "/home/anaconda3/envs/pytorch/lib/python3.8/site-packages/fastNLP/core/trainer.py", line 615, in train self._train() File "/home/anaconda3/envs/pytorch/lib/python3.8/site-packages/fastNLP/core/trainer.py", line 683, in _train self._grad_backward(loss) File "/home/anaconda3/envs/pytorch/lib/python3.8/site-packages/fastNLP/core/trainer.py", line 781, in _grad_backward loss.backward() File "/home/anaconda3/envs/pytorch/lib/python3.8/site-packages/torch/_tensor.py", line 255, in backward torch.autograd.backward(self, gradient, retain_graph, create_graph, inputs=inputs) File "/home/anaconda3/envs/pytorch/lib/python3.8/site-packages/torch/autograd/init.py", line 147, in backward Variable._execution_engine.run_backward( RuntimeError: Function MmBackward returned an invalid gradient at index 0 - got [712, 768] but expected shape compatible with [712, 21128] 我比较奇怪的是为什么forward没有报错而在backward报错了；以及想问下您，结果是在目前这个master版本下面跑出来的吗？有没有什么debug的头绪呢？ 感谢！

environment: pytorch==1.9.0+cuda111 transformers==4.8.2 fastNLP==0.6.0

[Onion12138]

我的版本能够运行，尝试打印一下tensor的维度。