fix attention.py validation bug

examples/wenetspeech/asr1/run.sh

@@ -3,10 +3,10 @@
 . path.sh || exit 1;
 set -e
 
-gpus=0,1,2,3,4,5,6,7
-stage=0
+gpus=1
+stage=3
 stop_stage=100
-conf_path=conf/conformer.yaml
+conf_path=conf/bitransformer_decoder_conformer.yaml
 ips=  #xxx.xxx.xxx.xxx,xxx.xxx.xxx.xxx
 decode_conf_path=conf/tuning/decode.yaml
 average_checkpoint=true

paddlespeech/s2t/modules/attention.py

@@ -101,10 +101,7 @@ class MultiHeadedAttention(nn.Layer):
             if self.train_stage:
                 self.build_kv()
                 self.train_stage = False
-            weight = paddle.concat(
-                [self.linear_k.weight, self.linear_v.weight], axis=-1)
-            bias = paddle.concat([self.linear_k.bias, self.linear_v.bias])
-            k, v = F.linear(key, weight, bias).view(n_batch, -1, 2 * self.h,
+            k, v = F.linear(key, self.weight, self.bias).view(n_batch, -1, 2 * self.h,
                                                     self.d_k).split(
                                                         2, axis=2)