Merge branch 'main' of https://github.com/microsoft/onnxruntime-genai …

…into baijumeswani/decoder-pipeline

.pipelines/macos-ios-cocoapods-publishing.yml

@@ -0,0 +1,22 @@
+parameters:
+- name: ort_version
+  displayName: 'OnnxRuntime version'
+  type: string
+  default: '1.20.0'
+
+- name: build_config
+  displayName: 'Build Configuration'
+  type: string
+  default: 'Release'
+  values:
+  - 'Debug'
+  - 'RelWithDebInfo'
+  - 'Release'
+  - 'MinSizeRel'
+
+trigger: none
+stages:
+- template: stages/macos-ios-cocoapods-packaging-stage.yml
+  parameters:
+    ort_version: ${{ parameters.ort_version }}
+    build_config: ${{ parameters.build_config }}

.pipelines/stages/jobs/capi-packaging-job.yml

@@ -111,5 +111,6 @@ jobs:
     - template: steps/capi-appleframework-step.yml
       parameters:
         build_config: ${{ parameters.build_config }}
+        build_settings_file: "tools/ci_build/github/apple/default_full_ios_framework_build_settings.json"
 
   - template: steps/compliant-and-cleanup-step.yml

.pipelines/stages/jobs/macos-ios-cocoapods-packaging-job.yml

@@ -0,0 +1,68 @@
+parameters:
+- name: build_config
+  type: string
+- name: ort_version
+  type: string
+
+jobs:
+- job: macos_ios_cocoapods_packaging
+  pool:
+    vmImage: 'macOS-latest'
+  variables:
+    buildSettingsFile: "tools/ci_build/github/apple/default_full_apple_framework_build_settings.json"
+    cPodName: onnxruntime-genai-c
+    objcPodName: onnxruntime-genai-objc
+
+  workspace:
+    clean: all
+  steps:
+    - checkout: self
+      clean: true
+      submodules: none
+
+    - template: steps/utils/set-genai-version.yml
+    - template: steps/utils/set-cmake-build-type.yml
+      parameters:
+        build_config: ${{parameters.build_config}}
+
+
+    - task: CmdLine@2
+      displayName: Build MacOS & iOS CocoaPods Packages
+      inputs:
+        script: |
+          set -e -x
+          python3 -m pip install requests huggingface_hub
+          python3 tools/ci_build/github/apple/build_and_assemble_apple_pods.py \
+            --build-dir "$(Build.BinariesDirectory)/apple_framework" \
+            --staging-dir "$(Build.BinariesDirectory)/staging" \
+            --pod-version "$(genai_version)" \
+            --test \
+            --variant Full \
+            --build-settings-file "${{ variables.buildSettingsFile }}" \
+            --build-apple-framework-arg=--config=$(cmake_build_type) \
+            --ort-version ${{parameters.ort_version}}
+
+    - script: |
+        set -e -x
+
+        for POD_NAME in "${{ variables.cPodName}}" "${{ variables.objcPodName }}";
+        do
+          ./tools/ci_build/github/apple/assemble_apple_packaging_artifacts.sh \
+            "$(Build.BinariesDirectory)/staging" \
+            "$(Build.ArtifactStagingDirectory)" \
+            "${POD_NAME}" \
+            "$(genai_version)"
+        done
+
+        # copy over helper script for use in release pipeline
+        # cp tools/ci_build/github/apple/package_release_tasks.py "$(Build.ArtifactStagingDirectory)"
+      displayName: "Assemble artifacts"
+
+    - script: |
+        set -e -x
+        ls -R "$(Build.ArtifactStagingDirectory)"
+      displayName: "List staged artifacts"
+
+    - publish: "$(Build.ArtifactStagingDirectory)"
+      artifact: "macos_ios_cocoapods_packaging_artifacts"
+      displayName: "Publish artifacts"

.pipelines/stages/jobs/nuget-packaging-job.yml

@@ -179,6 +179,8 @@ jobs:
 
   - template: steps/utils/set-genai-version.yml
 
+  - task: NuGetAuthenticate@1
+
   - powershell: |
       dotnet --info
       dotnet build Microsoft.ML.OnnxRuntimeGenAI.csproj -p:Configuration="$(buildConfig)" -p:IncludeMobileTargets=true --verbosity normal

.pipelines/stages/jobs/steps/capi-appleframework-step.yml

@@ -2,7 +2,8 @@ parameters:
 - name: build_config
   type: string
   default: 'release'
-
+- name: build_settings_file
+  type: string
 steps:
 
 - checkout: self
@@ -11,6 +12,9 @@ steps:
   submodules: recursive
 
 - template: utils/set-genai-version.yml
+- template: utils/set-cmake-build-type.yml
+  parameters:
+    build_config: ${{parameters.build_config}}
 
 - template: utils/set-nightly-build-option-variable.yml
 
@@ -19,7 +23,8 @@ steps:
     python3 -m pip install requests
     python3 tools/ci_build/github/apple/build_apple_framework.py \
           --build_dir "$(Build.BinariesDirectory)/apple_framework" \
-          tools/ci_build/github/apple/default_full_ios_framework_build_settings.json
+          --config $(cmake_build_type) \
+          "${{ parameters.build_settings_file }}"
 
     mkdir $(Build.BinariesDirectory)/artifacts
     mkdir -p $(Build.BinariesDirectory)/artifacts_staging/onnxruntime-genai.xcframework

.pipelines/stages/macos-ios-cocoapods-packaging-stage.yml

@@ -0,0 +1,15 @@
+parameters:
+- name: ort_version
+  type: string
+
+- name: build_config
+  type: string
+  default: 'Release'
+
+stages:
+- stage: Build_MacOS_iOS_CocoaPods_Archive
+  jobs:
+  - template: jobs/macos-ios-cocoapods-packaging-job.yml
+    parameters:
+      ort_version: ${{ parameters.ort_version }}
+      build_config: ${{ parameters.build_config }}

README.md

@@ -1,5 +1,8 @@
 # ONNX Runtime generate() API
 
+## *Main branch contains new API changes and examples in main branch reflect these changes. For example scripts compatible with current release (0.5.2), [see release branch](https://github.com/microsoft/onnxruntime-genai/tree/rel-0.5.2).*
+
+
 [![Latest version](https://img.shields.io/nuget/vpre/Microsoft.ML.OnnxRuntimeGenAI.Managed?label=latest)](https://www.nuget.org/packages/Microsoft.ML.OnnxRuntimeGenAI.Managed/absoluteLatest)
 
 Run Llama, Phi, Gemma, Mistral with ONNX Runtime.
@@ -50,8 +53,6 @@ See https://onnxruntime.ai/docs/genai/howto/install
 
 3. Run the model
 
-   *Main branch contains new API changes. For examples compatible with current release (0.5.2), [see release branch](https://github.com/microsoft/onnxruntime-genai/tree/rel-0.5.2).*
-
    ### Build from source / Next release (0.6.0)
 
    ```python

build.py

@@ -229,13 +229,11 @@ def _validate_build_dir(args: argparse.Namespace):
             # also tweak build directory name for mac builds
             target_sys = "macOS"
 
-        # set to a config specific build dir if no build_dir specified from command arguments
-        args.build_dir = Path("build") / target_sys / args.config
+        args.build_dir = Path("build") / target_sys
 
+    # set to a config specific build dir. it should exist unless we're creating the cmake setup
     is_strict = not args.update
-    # Use user-specified build_dir and ignore args.config
-    # This is to better accommodate the existing cmake presets which can uses arbitrary paths.
-    args.build_dir = args.build_dir.resolve(strict=is_strict)
+    args.build_dir = args.build_dir.resolve(strict=is_strict) / args.config
 
 
 def _validate_cuda_args(args: argparse.Namespace):
@@ -505,6 +503,7 @@ def update(args: argparse.Namespace, env: dict[str, str]):
             "-DENABLE_PYTHON=OFF",
             "-DENABLE_TESTS=OFF",
             "-DENABLE_MODEL_BENCHMARK=OFF",
+            "-DCMAKE_OSX_DEPLOYMENT_TARGET=" + args.apple_deploy_target,
             f"-DBUILD_APPLE_FRAMEWORK={'ON' if args.build_apple_framework else 'OFF'}",
             "-DPLATFORM_NAME=" + platform_name,
         ]

examples/c/src/phi3.cpp

@@ -164,19 +164,6 @@ void CXX_API(const char* model_path) {
           is_first_token = false;
         }
 
-        // Show usage of GetOutput
-        std::unique_ptr<OgaTensor> output_logits = generator->GetOutput("logits");
-
-        // Assuming output_logits.Type() is float as it's logits
-        // Assuming shape is 1 dimensional with shape[0] being the size
-        auto logits = reinterpret_cast<float*>(output_logits->Data());
-
-        // Print out the logits using the following snippet, if needed
-        //auto shape = output_logits->Shape();
-        //for (size_t i=0; i < shape[0]; i++)
-        //   std::cout << logits[i] << " ";
-        //std::cout << std::endl;
-
         const auto num_tokens = generator->GetSequenceCount(0);
         const auto new_token = generator->GetSequenceData(0)[num_tokens - 1];
         std::cout << tokenizer_stream->Decode(new_token) << std::flush;

src/models/decoder_only.h

@@ -24,7 +24,7 @@ struct DecoderOnly_State : State {
   void RewindTo(size_t index) override;
 
  private:
-  void UpdateInputsOutputs(DeviceSpan<int32_t>& next_tokens, DeviceSpan<int32_t> next_indices, int total_length);
+  void UpdateInputsOutputs(DeviceSpan<int32_t>& next_tokens, DeviceSpan<int32_t> beam_indices, int total_length);
 
   const DecoderOnly_Model& model_;
   CapturedGraphInfoPtr captured_graph_info_;

src/objectivec/cxx_api.h

@@ -0,0 +1,17 @@
+// Copyright (c) Microsoft Corporation. All rights reserved.
+// Licensed under the MIT License.
+
+// wrapper for ORT GenAI C/C++ API headers
+
+#if defined(__clang__)
+#pragma clang diagnostic push
+// ignore clang documentation-related warnings
+// instead, we will rely on Doxygen warnings for the C/C++ API headers
+#pragma clang diagnostic ignored "-Wdocumentation"
+#endif  // defined(__clang__)
+
+#include "ort_genai.h"
+
+#if defined(__clang__)
+#pragma clang diagnostic pop
+#endif  // defined(__clang__)

src/objectivec/error_utils.h

@@ -0,0 +1,29 @@
+// Copyright (c) Microsoft Corporation. All rights reserved.
+// Licensed under the MIT License.
+
+#import <Foundation/Foundation.h>
+
+#include <exception>
+
+#import "cxx_api.h"
+
+NS_ASSUME_NONNULL_BEGIN
+
+extern NSString* const kOgaErrorDomain;
+
+void OGASaveCodeAndDescriptionToError(int code, const char* description, NSError** error);
+void OGASaveCodeAndDescriptionToError(int code, NSString* description, NSError** error);
+void OGASaveExceptionToError(const std::exception& e, NSError** error);
+
+// helper macros to catch and handle C++ exceptions
+#define OGA_OBJC_API_IMPL_CATCH(error, failure_return_value) \
+  catch (const std::exception& e) {                          \
+    OGASaveExceptionToError(e, (error));                     \
+    return (failure_return_value);                           \
+  }
+
+#define OGA_OBJC_API_IMPL_CATCH_RETURNING_BOOL(error) OGA_OBJC_API_IMPL_CATCH(error, NO)
+
+#define OGA_OBJC_API_IMPL_CATCH_RETURNING_NULLABLE(error) OGA_OBJC_API_IMPL_CATCH(error, nil)
+
+NS_ASSUME_NONNULL_END

src/objectivec/error_utils.mm

@@ -0,0 +1,34 @@
+// Copyright (c) Microsoft Corporation. All rights reserved.
+// Licensed under the MIT License.
+
+#import "error_utils.h"
+
+NS_ASSUME_NONNULL_BEGIN
+
+NSString* const kOgaErrorDomain = @"onnxruntime-genai";
+const int kOgaErrorCode = 0x0A;
+
+void OGASaveCodeAndDescriptionToError(int code, const char* descriptionCstr, NSError** error) {
+  if (!error) return;
+
+  NSString* description = [NSString stringWithCString:descriptionCstr
+                                             encoding:NSUTF8StringEncoding];
+
+  *error = [NSError errorWithDomain:kOgaErrorDomain
+                               code:code
+                           userInfo:@{NSLocalizedDescriptionKey : description}];
+}
+
+void OGASaveCodeAndDescriptionToError(int code, NSString* description, NSError** error) {
+  if (!error) return;
+
+  *error = [NSError errorWithDomain:kOgaErrorDomain
+                               code:code
+                           userInfo:@{NSLocalizedDescriptionKey : description}];
+}
+
+void OGASaveExceptionToError(const std::exception& e, NSError** error) {
+  OGASaveCodeAndDescriptionToError(kOgaErrorCode, e.what(), error);
+}
+
+NS_ASSUME_NONNULL_END