diff --git a/.Jenkinsfile b/.Jenkinsfile
index 75156959..f2740e5b 100644
--- a/.Jenkinsfile
+++ b/.Jenkinsfile
@@ -46,7 +46,7 @@ pipeline {
                     agent {
                         docker {
                             image 'fnlp:torch-1.6'
-                            args '-u root:root -v ${JENKINS_HOME}/html/docs:/docs -v ${JENKINS_HOME}/html/_ci:/ci --gpus all'
+                            args '-u root:root -v ${JENKINS_HOME}/html/docs:/docs -v ${JENKINS_HOME}/html/_ci:/ci --gpus all --shm-size 1G'
                         }
                     }
                     steps {
@@ -62,7 +62,7 @@ pipeline {
                     agent {
                         docker {
                             image 'fnlp:paddle'
-                            args '-u root:root -v ${JENKINS_HOME}/html/docs:/docs -v ${JENKINS_HOME}/html/_ci:/ci --gpus all'
+                            args '-u root:root -v ${JENKINS_HOME}/html/docs:/docs -v ${JENKINS_HOME}/html/_ci:/ci --gpus all --shm-size 1G'
                         }
                     }
                     steps {
@@ -82,7 +82,7 @@ pipeline {
                 //     agent {
                 //         docker {
                 //             image 'fnlp:jittor'
-                //             args '-u root:root -v ${JENKINS_HOME}/html/docs:/docs -v ${JENKINS_HOME}/html/_ci:/ci --gpus all'
+                //             args '-u root:root -v ${JENKINS_HOME}/html/docs:/docs -v ${JENKINS_HOME}/html/_ci:/ci --gpus all --shm-size 1G'
                 //         }
                 //     }
                 //     steps {
diff --git a/tests/models/torch/test_seq2seq_model.py b/tests/models/torch/test_seq2seq_model.py
index ee1775d3..abae1f67 100755
--- a/tests/models/torch/test_seq2seq_model.py
+++ b/tests/models/torch/test_seq2seq_model.py
@@ -25,11 +25,11 @@ def prepare_env():
 
 
 def train_model(model, src_words_idx, tgt_words_idx, tgt_seq_len,  src_seq_len):
-    optimizer = optim.Adam(model.parameters(), lr=1e-2)
+    optimizer = optim.Adam(model.parameters(), lr=5e-3)
     mask = seq_len_to_mask(tgt_seq_len).eq(0)
     target = tgt_words_idx.masked_fill(mask, -100)
 
-    for i in range(100):
+    for i in range(50):
         optimizer.zero_grad()
         pred = model(src_words_idx, tgt_words_idx, src_seq_len)['pred']  # bsz x max_len x vocab_size
         loss = F.cross_entropy(pred.transpose(1, 2), target)