self_attn# core.nn.self_attn(module, *, x, rotary_time_emb, x_seqlen, impl)[source]# forward_varlen for RotarySelfAttention, on either backend. Return type: Tensor
