Add tests for mapping-related operators (#179)

* init

* finish vector_to_string

* add more test

Co-authored-by: Ze Tao <zetao@microsoft.com>

includes/ocos.h

@@ -56,6 +56,10 @@ struct OrtTensorDimensions : std::vector<int64_t> {
       s *= *it;
     return s;
   }
+
+  bool IsScalar() const{
+    return empty();
+  }
 };
 
 

onnxruntime_extensions/_cuops.py

@@ -53,7 +53,7 @@ class VectorToString(CustomOp):
 
     @classmethod
     def get_outputs(cls):
-        return [cls.io_def('text', onnx_proto.TensorProto.STRING, [])]
+        return [cls.io_def('text', onnx_proto.TensorProto.STRING, [None])]
 
     @classmethod
     def serialize_attr(cls, attrs):

operators/text/vector_to_string.cc

@@ -24,11 +24,11 @@ std::vector<std::string> VectorToStringImpl::Compute(const void* input, const Or
 
   const int64_t* ptr = static_cast<const int64_t*>(input);
 
-  if (vector_len_ == 1 && input_dim.size() == 1) {
+  if (vector_len_ == 1 && (input_dim.size() == 1 || input_dim.IsScalar())) {
     // only hit when the key is a scalar and the input is a vector
     output_dim = input_dim;
   } else {
-    if (input_dim[input_dim.size() - 1] != vector_len_) {
+    if (input_dim.IsScalar() || input_dim[input_dim.size() - 1] != vector_len_) {
       ORT_CXX_API_THROW(MakeString("Incompatible dimension: required vector length should be ", vector_len_), ORT_INVALID_ARGUMENT);
     }
 

test/shared_test/test_ortops_strings.cc

@@ -402,3 +402,159 @@ TEST(utils, test_string_join_dims_empty_values_scalar) {
 
   TestInference(*ort_env, model_path.c_str(), inputs, outputs, GetLibraryPath());
 }
+
+TEST(string_operator, test_vector_to_string) {
+  auto ort_env = std::make_unique<Ort::Env>(ORT_LOGGING_LEVEL_WARNING, "Default");
+
+  std::vector<TestValue> inputs(1);
+  inputs[0].name = "token_ids";
+  inputs[0].element_type = ONNXTensorElementDataType::ONNX_TENSOR_ELEMENT_DATA_TYPE_INT64;
+  inputs[0].dims = {5};
+  inputs[0].values_int64 = {0, 1, 2, 3, 4};
+
+  std::vector<TestValue> outputs(1);
+  outputs[0].name = "text";
+  outputs[0].element_type = ONNXTensorElementDataType::ONNX_TENSOR_ELEMENT_DATA_TYPE_STRING;
+  outputs[0].dims = {5};
+  outputs[0].values_string = {"a", "unk", "b", "c", "unk"};
+
+  std::filesystem::path model_path = __FILE__;
+  model_path = model_path.parent_path();
+  model_path /= "..";
+  model_path /= "data";
+  model_path /= "test_vector_to_string_scalar_map.onnx";
+  TestInference(*ort_env, model_path.c_str(), inputs, outputs, GetLibraryPath());
+
+  inputs[0].dims = {5};
+  inputs[0].values_int64 = {1000, 111, 2323, 444, 555};
+
+  outputs[0].dims = {5};
+  outputs[0].values_string = {"unk", "unk", "unk", "unk", "unk"};
+  TestInference(*ort_env, model_path.c_str(), inputs, outputs, GetLibraryPath());
+
+  inputs[0].dims = {0};
+  inputs[0].values_int64 = {};
+
+  outputs[0].dims = {0};
+  outputs[0].values_string = {};
+  TestInference(*ort_env, model_path.c_str(), inputs, outputs, GetLibraryPath());
+
+  inputs[0].dims = {5, 3};
+  inputs[0].values_int64 = {0, 0, 0, 0, 0, 1, 3, 0, 1, 100, 0, 1, 43, 23, 11};
+
+  outputs[0].dims = {5};
+  outputs[0].values_string = {"a", "b", "c", "unk", "unk"};
+
+  model_path = model_path.parent_path();
+  model_path /= "..";
+  model_path /= "data";
+  model_path /= "test_vector_to_string_vector_map.onnx";
+  TestInference(*ort_env, model_path.c_str(), inputs, outputs, GetLibraryPath());
+
+  inputs[0].dims = {0, 3};
+  inputs[0].values_int64 = {};
+
+  outputs[0].dims = {0};
+  outputs[0].values_string = {};
+  TestInference(*ort_env, model_path.c_str(), inputs, outputs, GetLibraryPath());
+
+  inputs[0].dims = {};
+  inputs[0].values_int64 = {111};
+
+  outputs[0].dims = {};
+  outputs[0].values_string = {"unk"};
+
+  model_path = model_path.parent_path();
+  model_path /= "..";
+  model_path /= "data";
+  model_path /= "test_vector_to_string_scalar_input.onnx";
+  TestInference(*ort_env, model_path.c_str(), inputs, outputs, GetLibraryPath());
+}
+
+TEST(string_operator, test_string_to_vector) {
+  auto ort_env = std::make_unique<Ort::Env>(ORT_LOGGING_LEVEL_WARNING, "Default");
+
+  std::vector<TestValue> inputs(1);
+  inputs[0].name = "text";
+  inputs[0].element_type = ONNXTensorElementDataType::ONNX_TENSOR_ELEMENT_DATA_TYPE_STRING;
+  inputs[0].dims = {5};
+  inputs[0].values_string = {"black", "white", "black", "中文", "英文"};
+
+  std::vector<TestValue> outputs(1);
+  outputs[0].name = "token_ids";
+  outputs[0].element_type = ONNXTensorElementDataType::ONNX_TENSOR_ELEMENT_DATA_TYPE_INT64;
+  outputs[0].dims = {5,3};
+  outputs[0].values_int64 = {0, 1, 2, 2, 3, 4, 0, 1, 2, 3, 4, 4, -1, -1, -1};
+
+
+  std::filesystem::path model_path = __FILE__;
+  model_path = model_path.parent_path();
+  model_path /= "..";
+  model_path /= "data";
+  model_path /= "test_string_to_vector.onnx";
+  TestInference(*ort_env, model_path.c_str(), inputs, outputs, GetLibraryPath());
+
+  inputs[0].dims = {0};
+  inputs[0].values_string = {};
+
+  outputs[0].dims = {0, 3};
+  outputs[0].values_int64 = {};
+  TestInference(*ort_env, model_path.c_str(), inputs, outputs, GetLibraryPath());
+
+  inputs[0].dims = {1};
+  inputs[0].values_string = {""};
+
+  outputs[0].dims = {1, 3};
+  outputs[0].values_int64 = {-1, -1, -1};
+  TestInference(*ort_env, model_path.c_str(), inputs, outputs, GetLibraryPath());
+
+  inputs[0].dims = {};
+  inputs[0].values_string = {""};
+
+  outputs[0].dims = {3};
+  outputs[0].values_int64 = {-1, -1, -1};
+
+  model_path = model_path.parent_path();
+  model_path /= "..";
+  model_path /= "data";
+  model_path /= "test_string_to_vector_scalar.onnx";
+  TestInference(*ort_env, model_path.c_str(), inputs, outputs, GetLibraryPath());
+}
+
+TEST(string_operator, test_string_mapping) {
+  auto ort_env = std::make_unique<Ort::Env>(ORT_LOGGING_LEVEL_WARNING, "Default");
+
+  std::vector<TestValue> inputs(1);
+  inputs[0].name = "input";
+  inputs[0].element_type = ONNXTensorElementDataType::ONNX_TENSOR_ELEMENT_DATA_TYPE_STRING;
+  inputs[0].dims = {5};
+  inputs[0].values_string = {"Orange and Yellow", "不知道啥颜色", "No color", "black", "white"};
+
+  std::vector<TestValue> outputs(1);
+  outputs[0].name = "output";
+  outputs[0].element_type = ONNXTensorElementDataType::ONNX_TENSOR_ELEMENT_DATA_TYPE_STRING;
+  outputs[0].dims = {5};
+  outputs[0].values_string = {"Maybe", "也不知道可不可以", "No color", "OK", "Not OK"};
+
+
+  std::filesystem::path model_path = __FILE__;
+  model_path = model_path.parent_path();
+  model_path /= "..";
+  model_path /= "data";
+  model_path /= "test_string_mapping.onnx";
+  TestInference(*ort_env, model_path.c_str(), inputs, outputs, GetLibraryPath());
+
+  inputs[0].dims = {};
+  inputs[0].values_string = {"不知道啥颜色"};
+
+  outputs[0].dims = {};
+  outputs[0].values_string = {"也不知道可不可以"};
+  TestInference(*ort_env, model_path.c_str(), inputs, outputs, GetLibraryPath());
+
+  inputs[0].dims = {};
+  inputs[0].values_string = {""};
+
+  outputs[0].dims = {};
+  outputs[0].values_string = {""};
+  TestInference(*ort_env, model_path.c_str(), inputs, outputs, GetLibraryPath());
+}