funasr/models/sense_voice/decoder.py
@@ -360,6 +360,7 @@ """Score.""" ys_mask = subsequent_mask(len(ys), device=x.device).unsqueeze(0) logp = self.forward(ys.unsqueeze(0), x.unsqueeze(0), cache=state) logp = torch.log_softmax(logp, dim=-1) return logp.squeeze(0)[-1, :], state
