I meet a bug when I try to use DIFFormer module for a heterograph data, where the first dimension of Q (denoted by N) and K (denoted by L) are different. A data dimension conflict was encountered on line 29 of the full_attention_conv function:

## difformer.py [line 29]
attention_num += vs_sum.unsqueeze(0).repeat(vs.shape[0], 1, 1) # [N, H, D]

The vs.shape[0] should be qs.shape[0] or N to fix the bug.