diff --git a/tf_adapter/kernels/geop_npu.cc b/tf_adapter/kernels/geop_npu.cc
index 87dd62d71512d8e34789944d324956d73f306c5f..d866b67531732191cccfd31950a0b5f592aca46e 100644
--- a/tf_adapter/kernels/geop_npu.cc
+++ b/tf_adapter/kernels/geop_npu.cc
@@ -1734,9 +1734,19 @@ Status GeOp::BuildInputTensorInfo(OpKernelContext *const ctx, std::vector<Tensor
   int num_inputs = ctx->num_inputs();
   std::string cur_input_shapes;
 
+  input_shapes_vec_.resize(num_inputs, TensorShape());
+
+  if (init_options_["ge.jit_compile"] == "0") {
+    FuzzAllShape(ctx);
+  } else {
+    MaybeUpdateShape(ctx);
+  }
+
   // populate inputs
   for (int i = 0; i < num_inputs; i++) {
     Tensor tensor(ctx->input(i));
+    ADP_LOG(INFO) << "Refresh input tensor " << i << " tensor1 to " << tensor.DebugString();
+
     bool is_equal = false;
     if (GraphCheckInputEqualConstOp(tensor, i, is_equal) != Status::OK()) {
       return errors::Internal("Const op value not equal with tensor :", i);
@@ -1786,16 +1796,9 @@ Status GeOp::BuildInputTensorInfo(OpKernelContext *const ctx, std::vector<Tensor
       cur_input_shapes += input_shape;
     }
     inputs.push_back(input);
-    input_vec.push_back(tensor);
-  }
-  if (sess_options_["ge.inputShape"].empty()) {
-    if (!cur_input_shapes.empty() && input_shapes_.empty()) {
-      input_shapes_ = cur_input_shapes;
-    } else if (input_shapes_ != cur_input_shapes && dynamic_input_ != "1") {
-      return errors::Internal("The input shape of ", ctx->op_kernel().name(),
-                              " is dynamic, please ensure that npu option[dynamic_input] is set"
-                              " correctly, for more details please refer to the migration guide.");
-    }
+    Tensor tensor1 = Tensor(data_type, input_shapes_vec_[i]);
+    ADP_LOG(INFO) << "Refresh input tensor1 " << i << " tensor1 to " << tensor1.DebugString();
+    input_vec.push_back(tensor1);
   }
   return Status::OK();
 }
diff --git a/tf_adapter/kernels/geop_npu.h b/tf_adapter/kernels/geop_npu.h
index 50c38544a9c4c0c7a1ceaef7b02d51c9cfd6161d..385b43848427ba52a4655a1b85b5e048e1a2faa3 100644
--- a/tf_adapter/kernels/geop_npu.h
+++ b/tf_adapter/kernels/geop_npu.h
@@ -20,6 +20,7 @@
 #include <unordered_map>
 #include <atomic>
 
+#include "tf_adapter/common/adapter_logger.h"
 #include "tensorflow/core/common_runtime/function.h"
 #include "tensorflow/core/framework/op_kernel.h"
 #include "tensorflow/core/platform/mutex.h"
@@ -128,6 +129,65 @@ private:
 
   bool IsDynamicConfig();
 
+  void FuzzAllShape(OpKernelContext *const ctx) {
+    for (size_t i = 0UL; i < static_cast<size_t>(ctx->num_inputs()); i++) {
+      auto &value_shape = ctx->input(static_cast<int32_t>(i)).shape();
+      TensorShape shape;
+      std::vector<int64> dims;
+      ADP_LOG(INFO) << "value_shape dims " << value_shape.dims();
+      for (int j = 0; j < value_shape.dims(); j++) {
+        dims.push_back(-1);
+      }
+      ADP_LOG(INFO) << "dims size is " << dims.size();
+      for (auto dim : dims) {
+        ADP_LOG(INFO) << "dim is : " << dim;
+      }
+      auto status = TensorShapeUtils::MakeShape(dims.data(), static_cast<int32_t>(dims.size()), &shape);
+      ADP_LOG(INFO) << "FuzzAllShape Init input " << i << " shape " << shape.DebugString() << "status is"
+                    << status.ok();
+      input_shapes_vec_[i] = shape;
+    }
+  }
+
+  void MaybeUpdateShape(OpKernelContext *const ctx) {
+    for (size_t i = 0UL; i < static_cast<size_t>(ctx->num_inputs()); i++) {
+      auto &shape = input_shapes_vec_[i];
+      auto &value_shape = ctx->input(static_cast<int32_t>(i)).shape();
+      if (shape.dims() == 0) {
+        build_flag_ = false;
+        shape = value_shape;
+        ADP_LOG(INFO) << "Init input " << i << " shape " << shape.DebugString();
+      } else {
+        PartialTensorShape sub_shape = shape;
+        if (sub_shape.IsCompatibleWith(value_shape)) {
+          continue;
+        } else {
+          build_flag_ = false;
+          ADP_LOG(INFO) << "Compat input " << i << " shape " << shape.DebugString() << " vs. "
+                        << value_shape.DebugString();
+          shape = MakeCompatShape(shape, value_shape);
+          ADP_LOG(INFO) << "Refresh input " << i << " shape to " << shape.DebugString();
+        }
+      }
+    }
+  }
+
+ TensorShape MakeCompatShape(const TensorShape &a, const TensorShape &b) {
+    const static auto kUnknownRankShape = TensorShape();
+    if (a.dims() != b.dims()) {
+      return kUnknownRankShape;
+    }
+    TensorShape shape;
+    static constexpr int64 kUnknownDim = -1;
+    std::vector<int64> dims;
+    for (int i = 0; i < a.dims(); i++) {
+      dims.push_back((a.dim_size(i) != b.dim_size(i)) ? kUnknownDim : a.dim_size(i));
+    }
+    auto status = TensorShapeUtils::MakeShape(dims.data(), static_cast<int32_t>(dims.size()), &shape);
+    //NPU_LOG_IF_ERROR(status);
+    return status.ok() ? shape : kUnknownRankShape;
+  }
+
   static const std::string INPUT_DESC;
   static const std::string OUTPUT_DESC;
   static const std::string SERIALIZE_FORMAT;
@@ -180,6 +240,7 @@ private:
   std::string recompute_mode_;
   std::string enable_graph_parallel_;
   std::string graph_parallel_option_path_;
+  std::vector<TensorShape> input_shapes_vec_;
 
   SessionId session_id_;
   AoeInitializeFunc aoe_initialize_;