[fairseq] 报错:TypeError: _broadcast_coalesced(): incompatible function arguments
前言我通过复写了模型的state_dict方法,具体就是给其增加了dynamic_mask(字典类型,里面是tensor),allocated_neuron_num(整型)。defstate_dict(self,destination=None,prefix='',keep_vars=False):state_dict=super().state_dict(destination,prefix,k