update libtorch inference

维石

2024-06-03 fd0992af3d1a2d2d098b1fab24f67f8c4cece39d

 funasr/models/sanm/attention.py

@@ -780,7 +780,7 @@
        return q, k, v

    def forward_attention(self, value, scores, mask, ret_attn):
        scores = scores + mask
        scores = scores + mask.to(scores.device)

        self.attn = torch.softmax(scores, dim=-1)
        context_layer = torch.matmul(self.attn, value)  # (batch, head, time1, d_k)