Update C, C++, and C# examples

examples/c/src/phi3.cpp

@@ -6,6 +6,7 @@
 #include <iomanip>
 #include <iostream>
 #include <string>
+#include <cstring>
 #include "ort_genai.h"
 #include <thread>
 #include <csignal>
@@ -119,9 +120,22 @@ void signalHandlerWrapper(int signum) {
   catch_terminate.signalHandler(signum);
 }
 
-void CXX_API(const char* model_path) {
+void CXX_API(const char* model_path, const char* execution_provider) {
+  std::cout << "Creating config..." << std::endl;
+  auto config = OgaConfig::Create(model_path);
+
+  config->ClearProviders();
+  std::string provider(execution_provider);
+  if (provider.compare("cpu") != 0) {
+    config->AppendProvider(execution_provider);
+    if (provider.compare("cuda") == 0) {
+      config->SetProviderOption(execution_provider, "enable_cuda_graph", "0");
+    }
+  }
+
   std::cout << "Creating model..." << std::endl;
-  auto model = OgaModel::Create(model_path);
+  auto model = OgaModel::Create(*config);
+
   std::cout << "Creating tokenizer..." << std::endl;
   auto tokenizer = OgaTokenizer::Create(*model);
   auto tokenizer_stream = OgaTokenizerStream::Create(*tokenizer);
@@ -209,10 +223,22 @@ bool CheckIfSessionTerminated(OgaResult* result, OgaGenerator* generator) {
   return false;
 }
 
-void C_API(const char* model_path) {
+void C_API(const char* model_path, const char* execution_provider) {
+  OgaConfig* config;
+  std::cout << "Creating config..." << std::endl;
+  CheckResult(OgaCreateConfig(model_path, &config));
+
+  CheckResult(OgaConfigClearProviders(&config));
+  if (strcmp(execution_provider, "cpu") != 0) {
+    CheckResult(OgaConfigAppendProvider(&config, execution_provider));
+    if (strcmp(execution_provider, "cuda") == 0) {
+      CheckResult(OgaConfigSetProviderOption(&config, execution_provider, "enable_cuda_graph", "0"));
+    }
+  }
+
   OgaModel* model;
   std::cout << "Creating model..." << std::endl;
-  OgaCreateModel(model_path, &model);
+  CheckResult(OgaCreateModelFromConfig(&config, &model));
 
   OgaTokenizer* tokenizer;
   std::cout << "Creating tokenizer..." << std::endl;
@@ -293,11 +319,13 @@ void C_API(const char* model_path) {
 }
 
 static void print_usage(int /*argc*/, char** argv) {
-  std::cerr << "usage: " << argv[0] << " model_path" << std::endl;
+  std::cerr << "usage: " << argv[0] << std::endl;
+  std::cerr << "model_path = " << argv[1] << std::endl;
+  std::cerr << "execution_provider = " << argv[2] << std::endl;
 }
 
 int main(int argc, char** argv) {
-  if (argc != 2) {
+  if (argc != 3) {
     print_usage(argc, argv);
     return -1;
   }
@@ -311,10 +339,10 @@ int main(int argc, char** argv) {
 
 #ifdef USE_CXX
   std::cout << "C++ API" << std::endl;
-  CXX_API(argv[1]);
+  CXX_API(argv[1], argv[2]);
 #else
   std::cout << "C API" << std::endl;
-  C_API(argv[1]);
+  C_API(argv[1], argv[2]);
 #endif
 
   return 0;

examples/c/src/phi3v.cpp

@@ -23,9 +23,22 @@ std::string trim(const std::string& str) {
 
 // C++ API Example
 
-void CXX_API(const char* model_path) {
+void CXX_API(const char* model_path, const char* execution_provider) {
+  std::cout << "Creating config..." << std::endl;
+  auto config = OgaConfig::Create(model_path);
+
+  config->ClearProviders();
+  std::string provider(execution_provider);
+  if (provider.compare("cpu") != 0) {
+    config->AppendProvider(execution_provider);
+    if (provider.compare("cuda") == 0) {
+      config->SetProviderOption(execution_provider, "enable_cuda_graph", "0");
+    }
+  }
+
   std::cout << "Creating model..." << std::endl;
-  auto model = OgaModel::Create(model_path);
+  auto model = OgaModel::Create(*config);
+
   std::cout << "Creating multimodal processor..." << std::endl;
   auto processor = OgaMultiModalProcessor::Create(*model);
 
@@ -98,10 +111,22 @@ void CheckResult(OgaResult* result) {
   }
 }
 
-void C_API(const char* model_path) {
+void C_API(const char* model_path, const char* execution_provider) {
+  OgaConfig* config;
+  std::cout << "Creating config..." << std::endl;
+  CheckResult(OgaCreateConfig(model_path, &config));
+
+  CheckResult(OgaConfigClearProviders(&config));
+  if (strcmp(execution_provider, "cpu") != 0) {
+    CheckResult(OgaConfigAppendProvider(&config, execution_provider));
+    if (strcmp(execution_provider, "cuda") == 0) {
+      CheckResult(OgaConfigSetProviderOption(&config, execution_provider, "enable_cuda_graph", "0"));
+    }
+  }
+
   OgaModel* model;
   std::cout << "Creating model..." << std::endl;
-  CheckResult(OgaCreateModel(model_path, &model));
+  CheckResult(OgaCreateModelFromConfig(&config, &model));
 
   OgaMultiModalProcessor* processor;
   std::cout << "Creating multimodal processor..." << std::endl;
@@ -184,11 +209,13 @@ void C_API(const char* model_path) {
 }
 
 static void print_usage(int /*argc*/, char** argv) {
-  std::cerr << "usage: " << argv[0] << " <model_path>" << std::endl;
+  std::cerr << "usage: " << argv[0] << std::endl;
+  std::cerr << "model_path = " << argv[1] << std::endl;
+  std::cerr << "execution_provider = " << argv[2] << std::endl;
 }
 
 int main(int argc, char** argv) {
-  if (argc != 2) {
+  if (argc != 3) {
     print_usage(argc, argv);
     return -1;
   }
@@ -199,10 +226,10 @@ int main(int argc, char** argv) {
 
 #ifdef USE_CXX
   std::cout << "C++ API" << std::endl;
-  CXX_API(argv[1]);
+  CXX_API(argv[1], argv[2]);
 #else
   std::cout << "C API" << std::endl;
-  C_API(argv[1]);
+  C_API(argv[1], argv[2]);
 #endif
 
   return 0;

examples/csharp/HelloPhi/Program.cs

@@ -8,6 +8,8 @@ void PrintUsage()
     Console.WriteLine("Usage:");
     Console.WriteLine("  -m model_path");
     Console.WriteLine("\t\t\t\tPath to the model");
+    Console.WriteLine("  -e execution_provider");
+    Console.WriteLine("\t\t\t\tExecution provider to run the model");
     Console.WriteLine("  --non-interactive (optional)");
     Console.WriteLine("\t\t\t\tInteractive mode");
 }
@@ -22,6 +24,7 @@ void PrintUsage()
 
 bool interactive = true;
 string modelPath = string.Empty;
+string executionProvider = string.Empty;
 
 uint i = 0;
 while (i < args.Length)
@@ -38,22 +41,42 @@ void PrintUsage()
             modelPath = Path.Combine(args[i+1]);
         }
     }
+    else if (arg == "-e")
+    {
+        if (i + 1 < args.Length)
+        {
+            executionProvider = Path.Combine(args[i+1]);
+        }
+    }
     i++;
 }
 
 if (string.IsNullOrEmpty(modelPath))
 {
     throw new Exception("Model path must be specified");
 }
+if (string.IsNullOrEmpty(executionProvider))
+{
+    throw new Exception("Execution provider must be specified");
+}
 
 Console.WriteLine("-------------");
 Console.WriteLine("Hello, Phi!");
 Console.WriteLine("-------------");
 
 Console.WriteLine("Model path: " + modelPath);
+Console.WriteLine("Execution provider: " + executionProvider);
 Console.WriteLine("Interactive: " + interactive);
 
-using Model model = new Model(modelPath);
+using Config config = new Config(modelPath);
+config.ClearProviders();
+if (!config.equals("cpu")) {
+    config.AppendProvider(executionProvider);
+    if (executionProvider.equals("cuda")) {
+        config.SetProviderOption(executionProvider, "enable_cuda_graph", "0");
+    }
+}
+using Model model = new Model(config);
 using Tokenizer tokenizer = new Tokenizer(model);
 
 var option = 2;

examples/csharp/HelloPhi/README.md

@@ -2,18 +2,22 @@
 
 ## Obtain a model
 
-You can download a published model from HuggingFace. For example, this is Phi-3 mini optimized for CPU and mobile. You can find other models here: 
+You can download a published model from Hugging Face. For example, this is Phi-3.5 mini optimized for CPU and mobile. You can find other models here: 
 
 ```script
-huggingface-cli download microsoft/Phi-3-mini-4k-instruct-onnx --include cpu_and_mobile/cpu-int4-rtn-block-32-acc-level-4/* --local-dir models
+huggingface-cli download microsoft/Phi-3.5-mini-instruct-onnx --include cpu_and_mobile/cpu-int4-rtn-block-32-acc-level-4/* --local-dir models
 move models\cpu_and_mobile\cpu-int4-rtn-block-32-acc-level-4 models\phi-3
 ```
 
-Alternatively you can build a model yourself using the model builder tool. See [Generate models using Model Builder](https://onnxruntime.ai/docs/genai/howto/build-model.html) for more details.
+Alternatively you can build a model yourself using the model builder. See [here](https://github.com/microsoft/onnxruntime-genai/blob/main/src/python/py/models/README.md) for more details.
 
 
 ## Run the model
 
 Open [HelloPhi.sln](HelloPhi.sln) and run the console application.
 
-Note that this application does not add a template to the prompt that you enter. If your model needs a template (e.g. `<|user|>\n{input} <|end|>\n<|assistant|>` for Phi-3) then please add this to your prompt.
+Notes:
+
+1. The `executionProvider` must be one of the following: `cpu`, `cuda`, or `dml`.
+
+2. This application does not add a template to the prompt that you enter. If your model needs a template (e.g. `<|user|>\n{input} <|end|>\n<|assistant|>` for Phi-3.5) then please add this to your prompt.

examples/csharp/HelloPhi3V/Program.cs

@@ -33,6 +33,8 @@ void PrintUsage()
     Console.WriteLine("Usage:");
     Console.WriteLine("  -m model_path");
     Console.WriteLine("\t\t\t\tPath to the model");
+    Console.WriteLine("  -e execution_provider");
+    Console.WriteLine("\t\t\t\tExecution provider for the model");
     Console.WriteLine("  --image_paths");
     Console.WriteLine("\t\t\t\tPath to the images");
     Console.WriteLine("  --non-interactive (optional), mainly for CI usage");
@@ -49,6 +51,7 @@ void PrintUsage()
 
 bool interactive = true;
 string modelPath = string.Empty;
+string executionProvider = string.Empty;
 List<string> imagePaths = new List<string>();
 
 uint i_arg = 0;
@@ -66,6 +69,13 @@ void PrintUsage()
             modelPath = Path.Combine(args[i_arg+1]);
         }
     }
+    else if (arg == "-e")
+    {
+        if (i_arg + 1 < args.Length)
+        {
+            executionProvider = Path.Combine(args[i_arg+1]);
+        }
+    }
     else if (arg == "--image_paths")
     {
         if (i_arg + 1 < args.Length)
@@ -76,14 +86,32 @@ void PrintUsage()
     i_arg++;
 }
 
+if (string.IsNullOrEmpty(modelPath))
+{
+    throw new Exception("Model path must be specified");
+}
+if (string.IsNullOrEmpty(executionProvider))
+{
+    throw new Exception("Execution provider must be specified");
+}
+
 Console.WriteLine("--------------------");
 Console.WriteLine("Hello, Phi-3-Vision!");
 Console.WriteLine("--------------------");
 
 Console.WriteLine("Model path: " + modelPath);
+Console.WriteLine("Execution provider: " + executionProvider);
 Console.WriteLine("Interactive: " + interactive);
 
-using Model model = new Model(modelPath);
+using Config config = new Config(modelPath);
+config.ClearProviders();
+if (!config.equals("cpu")) {
+    config.AppendProvider(executionProvider);
+    if (executionProvider.equals("cuda")) {
+        config.SetProviderOption(executionProvider, "enable_cuda_graph", "0");
+    }
+}
+using Model model = new Model(config);
 using MultiModalProcessor processor = new MultiModalProcessor(model);
 using var tokenizerStream = processor.CreateStream();