diff --git a/lite/core/optimizer/mir/fusion/__xpu__multi_encoder_slice_link_fuse_pass.cc b/lite/core/optimizer/mir/fusion/__xpu__multi_encoder_slice_link_fuse_pass.cc
index 7037805c28f..03948c51cf2 100644
--- a/lite/core/optimizer/mir/fusion/__xpu__multi_encoder_slice_link_fuse_pass.cc
+++ b/lite/core/optimizer/mir/fusion/__xpu__multi_encoder_slice_link_fuse_pass.cc
@@ -36,21 +36,22 @@ class XPUMultiEncoderSliceLinkFuser : public FuseBase {
     PMNode* layer_norm = nullptr;
     PMNode* layer_norm_out = nullptr;
 
-    auto* slice = OpNode("slice", "slice")
-                      ->assert_op_attr_satisfied<std::vector<int>>(
-                          "axes",
-                          [](const std::vector<int>& attr) {
-                            return attr.size() == 1 && attr[0] == 1;
-                          })
-                      ->assert_op_attr_satisfied<std::vector<int>>(
-                          "starts",
-                          [](const std::vector<int>& attr) {
-                            return attr.size() == 1 && attr[0] == 0;
-                          })
-                      ->assert_op_attr_satisfied<std::vector<int>>(
-                          "ends", [](const std::vector<int>& attr) {
-                            return attr.size() == 1 && attr[0] == 1;
-                          });
+    auto* slice =
+        OpNode("slice", "slice")
+            ->assert_op_attr_satisfied<std::vector<int>>(
+                "axes",
+                [](const std::vector<int>& attr) {
+                  return attr.size() == 1 && attr[0] == 1;
+                })
+            ->assert_op_attr_satisfied<std::vector<int>>(
+                "starts",
+                [](const std::vector<int>& attr) {
+                  return attr.size() == 1 && attr[0] == 0;
+                })
+            ->assert_op_attr_satisfied<std::vector<int>>(
+                "ends", [](const std::vector<int>& attr) {
+                  return attr.size() == 1 && attr[0] > 0 && attr[0] <= 20;
+                });
     if (pre_ln_) {
       xpu_encoder->assert_op_attr<bool>("norm_before", true);
       encoder_out->assert_is_op_input("layer_norm", "X");
diff --git a/lite/kernels/xpu/__xpu__multi_encoder_compute.cc b/lite/kernels/xpu/__xpu__multi_encoder_compute.cc
index c87a3e270d8..051372a2938 100644
--- a/lite/kernels/xpu/__xpu__multi_encoder_compute.cc
+++ b/lite/kernels/xpu/__xpu__multi_encoder_compute.cc
@@ -239,9 +239,10 @@ void XPUMultiEncoderCompute::PrepareForRun() {
   }
   // prepare with sice
   if ((param.slice_starts.size() > 0 && param.slice_starts[0] == 0) &&
-      (param.slice_ends.size() > 0 && param.slice_ends[0] == 1) &&
+      (param.slice_ends.size() > 0 && param.slice_ends[0] > 0 &&
+       param.slice_ends[0] <= 20) &&
       (param.slice_axes.size() > 0 && param.slice_axes[0] == 1)) {
-    slice_idx = 0;
+    slice_idx = param.slice_ends[0] - 1;
   }
   // prepare input_cast and output_cast guard_
   cast_in_guard_ = TargetWrapperXPU::MallocScratchPad(4 * 1024 * 1024);
diff --git a/lite/operators/__xpu__multi_encoder_op.cc b/lite/operators/__xpu__multi_encoder_op.cc
index 8ef7da2e7b6..11b027ed94a 100644
--- a/lite/operators/__xpu__multi_encoder_op.cc
+++ b/lite/operators/__xpu__multi_encoder_op.cc
@@ -42,9 +42,11 @@ bool XPUMultiEncoderOp::InferShapeImpl() const {
     seq_len = param_.PadSeqLen->data<int>()[0];
   }
   if ((param_.slice_starts.size() > 0 && param_.slice_starts[0] == 0) &&
-      (param_.slice_ends.size() > 0 && param_.slice_ends[0] == 1) &&
+      (param_.slice_ends.size() > 0 && param_.slice_ends[0] > 0 &&
+       param_.slice_ends[0] <= 20) &&
       (param_.slice_axes.size() > 0 && param_.slice_axes[0] == 1)) {
-    DDim out_dims(std::vector<int64_t>({batch_size, 1, head_num}));
+    DDim out_dims(
+        std::vector<int64_t>({batch_size, param_.slice_ends[0], head_num}));
     if (param_.slice_decrease_axis.size() > 0) {
       std::vector<int64_t> new_out_shape;
       for (size_t i = 0; i < slice_decrease_axis.size(); ++i) {
@@ -65,7 +67,7 @@ bool XPUMultiEncoderOp::InferShapeImpl() const {
       out_dims = new_dims;
     }
     if (param_.norm_before) {
-      param_.output->Resize({batch_size, 1, head_num});
+      param_.output->Resize({batch_size, param_.slice_ends[0], head_num});
     } else {
       param_.output->Resize(out_dims);
     }