修复BiLSTM中ht, ct状态在有seq_len时顺序错乱的bug

fastNLP/modules/encoder/lstm.py

@@ -56,8 +56,8 @@ class LSTM(nn.Module):
         :param seq_len: [batch, ] 序列长度, 若为 ``None``, 所有输入看做一样长. Default: ``None``
         :param h0: [batch, hidden_size] 初始隐状态, 若为 ``None`` , 设为全0向量. Default: ``None``
         :param c0: [batch, hidden_size] 初始Cell状态, 若为 ``None`` , 设为全0向量. Default: ``None``
-        :return (output, ht) 或 output: 若 ``get_hidden=True`` [batch, seq_len, hidden_size*num_direction] 输出序列
-            和 [batch, hidden_size*num_direction] 最后时刻隐状态.
+        :return (output, (ht, ct)): output: [batch, seq_len, hidden_size*num_direction] 输出序列
+            和 ht,ct: [num_layers*num_direction, batch, hidden_size] 最后时刻隐状态.
         """
         batch_size, max_len, _ = x.size()
         if h0 is not None and c0 is not None:
@@ -78,6 +78,7 @@ class LSTM(nn.Module):
                 output = output[unsort_idx]
             else:
                 output = output[:, unsort_idx]
+            hx = hx[0][:, unsort_idx], hx[1][:, unsort_idx]
         else:
             output, hx = self.lstm(x, hx)
         return output, hx