Merge branch 'master' into wenyi5608-greedy-sampling

.clang-format

@@ -0,0 +1,28 @@
+BasedOnStyle: Google
+IndentWidth: 4
+UseTab: Never
+ColumnLimit: 120
+
+Language: Cpp
+Standard: Cpp11
+
+AccessModifierOffset: -4
+AlignConsecutiveMacros: true
+AllowAllArgumentsOnNextLine: false
+AllowAllConstructorInitializersOnNextLine: false
+AllowAllParametersOfDeclarationOnNextLine: false
+AllowShortFunctionsOnASingleLine: Empty
+AllowShortIfStatementsOnASingleLine: Never
+AllowShortLambdasOnASingleLine: Empty
+AllowShortLoopsOnASingleLine: false
+AlwaysBreakBeforeMultilineStrings: false
+BinPackArguments: false
+BinPackParameters: false
+CommentPragmas: '^#'
+DerivePointerAlignment: false
+FixNamespaceComments: true
+IndentCaseLabels: false
+IndentPPDirectives: AfterHash
+ForEachMacros:
+  - foreach
+  - FOREACH_CHILD

.github/ISSUE_TEMPLATE/good_first_issue.yml

@@ -43,6 +43,7 @@ body:
       value: |
           - [Contribution guide - start here!](https://github.com/openvinotoolkit/openvino/blob/master/CONTRIBUTING.md)
           - [Intel DevHub Discord channel](https://discord.gg/7pVRxUwdWG) - engage in discussions, ask questions and talk to OpenVINO developers
+          - [How to link your Pull Request to an issue](https://docs.github.com/en/issues/tracking-your-work-with-issues/linking-a-pull-request-to-an-issue#manually-linking-a-pull-request-to-an-issue-using-the-pull-request-sidebar)
     validations:
       required: true
 

.github/workflows/llm_bench-python.yml

@@ -73,7 +73,7 @@ jobs:
           python-version: 3.8
       - name: Test stateful
         run: |
-          python -m pip install -r llm_bench/python/requirements.txt optimum
+          python -m pip install -r llm_bench/python/requirements.txt
           python -m pip uninstall --yes openvino
           python -m pip install openvino-nightly
           python llm_bench/python/convert.py --model_id TinyLlama/TinyLlama-1.1B-Chat-v1.0 --output_dir . --stateful

.github/workflows/stable_diffusion_1_5_cpp.yml

@@ -1,84 +1,115 @@
 name: stable_diffusion_1_5_cpp
+
 on:
   pull_request:
     paths:
       - image_generation/stable_diffusion_1_5/cpp/**
       - image_generation/common/**
       - .github/workflows/stable_diffusion_1_5_cpp.yml
       - thirdparty/openvino_tokenizers
+
+env:
+  working_directory: "./image_generation/stable_diffusion_1_5/cpp/"
+
 concurrency:
   group: ${{ github.workflow }}-${{ github.ref }}
   cancel-in-progress: true
+
 jobs:
   stable_diffusion_1_5_cpp-linux:
     runs-on: ubuntu-20.04-8-cores
+    defaults:
+      run:
+        # Do not ignore bash profile files. From:
+        # https://github.com/marketplace/actions/setup-miniconda#important
+        shell: bash -l {0}
     steps:
       - uses: actions/checkout@v4
         with:
           submodules: recursive
-      - uses: actions/setup-python@v4
+
+      - name: Setup conda
+        uses: conda-incubator/setup-miniconda@v3
         with:
-          python-version: 3.8
-      - name: Install OpenVINO
+          miniconda-version: "latest"
+          activate-environment: openvino_sd_cpp
+          python-version: "3.10"
+
+      - name: Install OpenVINO and other conda dependencies
         run: |
-          set -e
-          mkdir openvino
-          curl https://storage.openvinotoolkit.org/repositories/openvino/packages/nightly/2024.1.0-14645-e6dc0865128/l_openvino_toolkit_ubuntu20_2024.1.0.dev20240304_x86_64.tgz | tar --directory ./openvino/ --strip-components 1 -xz
-          sudo ./openvino/install_dependencies/install_openvino_dependencies.sh
-      - name: Download / convert models
+          conda activate openvino_sd_cpp
+          conda install -c conda-forge openvino c-compiler cxx-compiler make cmake
+          conda env config vars set LD_LIBRARY_PATH=$CONDA_PREFIX/lib:$LD_LIBRARY_PATH
+
+      - name: Install python dependencies
+        working-directory: ${{ env.working_directory }}
         run: |
-          set -e
-          source ./openvino/setupvars.sh
-          cd ./image_generation/stable_diffusion_1_5/cpp/scripts/
-          python -m pip install -U pip
-          python -m pip install -r ./requirements.txt
-          python -m pip install ../../../../thirdparty/openvino_tokenizers/
-          python convert_model.py -sd runwayml/stable-diffusion-v1-5 -b 1 -t FP16 -dyn True
+          conda activate openvino_sd_cpp
+          python -m pip install -r requirements.txt
+          python -m pip install ../../../thirdparty/openvino_tokenizers/[transformers]
+
+      - name: Download and convert model and tokenizer
+        working-directory: ${{ env.working_directory }}
+        run: |
+          conda activate openvino_sd_cpp
+          export MODEL_PATH="models/stable_diffusion_v1_5_ov/FP16"
+          optimum-cli export openvino --model runwayml/stable-diffusion-v1-5 --task stable-diffusion --convert-tokenizer --weight-format fp16 $MODEL_PATH
+          convert_tokenizer $MODEL_PATH/tokenizer/ --tokenizer-output-type i32 -o $MODEL_PATH/tokenizer/
+
       - name: Build app
+        working-directory: ${{ env.working_directory }}
         run: |
-          set -e
-          source ./openvino/setupvars.sh
-          cd ./image_generation/stable_diffusion_1_5/cpp/
+          conda activate openvino_sd_cpp
           cmake -DCMAKE_BUILD_TYPE=Release -S ./ -B ./build/
           cmake --build ./build/ --config Release --parallel
+
       - name: Run app
+        working-directory: ${{ env.working_directory }}
         run: |
-          set -e
-          source ./openvino/setupvars.sh
-          cd ./image_generation/stable_diffusion_1_5/cpp/build
-          ./stable_diffusion -m ../scripts/runwayml/stable-diffusion-v1-5 -t FP16_dyn
+          ./build/stable_diffusion -m ./models/stable_diffusion_v1_5_ov -t FP16
+
   stable_diffusion_1_5_cpp-windows:
       runs-on: windows-latest
       steps:
         - uses: actions/checkout@v4
           with:
             submodules: recursive
-        - uses: actions/setup-python@v4
+
+        - name: Setup conda
+          uses: conda-incubator/setup-miniconda@v3
           with:
-            python-version: 3.8
-        - name: Initialize OpenVINO
-          shell: cmd
+            miniconda-version: "latest"
+            activate-environment: openvino_sd_cpp
+            python-version: "3.10"
+
+        - name: Install OpenVINO and other conda dependencies
+          run: |
+            conda activate openvino_sd_cpp
+            conda install -c conda-forge openvino c-compiler cxx-compiler make cmake
+  
+        - name: Install python dependencies
+          working-directory: ${{ env.working_directory }}
           run: |
-            curl --output ov.zip https://storage.openvinotoolkit.org/repositories/openvino/packages/nightly/2024.1.0-14645-e6dc0865128/w_openvino_toolkit_windows_2024.1.0.dev20240304_x86_64.zip
-            unzip ov.zip
-        - name: Download / convert a model / tokenizer
-          shell: cmd
+            conda activate openvino_sd_cpp
+            python -m pip install -r requirements.txt
+            python -m pip install ../../../thirdparty/openvino_tokenizers/[transformers]
+  
+        - name: Download and convert model and tokenizer
+          working-directory: ${{ env.working_directory }}
           run: |
-            call w_openvino_toolkit_windows_2024.1.0.dev20240304_x86_64/setupvars.bat
-            cd ./image_generation/stable_diffusion_1_5/cpp/scripts/
-            python -m pip install -r ./requirements.txt
-            python -m pip install ../../../../thirdparty/openvino_tokenizers/
-            python convert_model.py -sd runwayml/stable-diffusion-v1-5 -b 1 -t FP16 -dyn True
+            conda activate openvino_sd_cpp
+            $env:MODEL_PATH='models/stable_diffusion_v1_5_ov/FP16'
+            optimum-cli export openvino --model runwayml/stable-diffusion-v1-5 --task stable-diffusion --convert-tokenizer --weight-format fp16 $env:MODEL_PATH
+            convert_tokenizer $env:MODEL_PATH/tokenizer/ --tokenizer-output-type i32 -o $env:MODEL_PATH/tokenizer/
+  
         - name: Build app
-          shell: cmd
+          working-directory: ${{ env.working_directory }}
           run: |
-            call w_openvino_toolkit_windows_2024.1.0.dev20240304_x86_64/setupvars.bat
-            cd ./image_generation/stable_diffusion_1_5/cpp/
+            conda activate openvino_sd_cpp
             cmake -DCMAKE_BUILD_TYPE=Release -S ./ -B ./build/
             cmake --build ./build/ --config Release --parallel
+  
         - name: Run app
-          shell: cmd
+          working-directory: ${{ env.working_directory }}
           run: |
-            call w_openvino_toolkit_windows_2024.1.0.dev20240304_x86_64/setupvars.bat
-            cd ./image_generation/stable_diffusion_1_5/cpp/build/
-            call "./Release/stable_diffusion.exe" -m ../scripts/runwayml/stable-diffusion-v1-5 -t FP16_dyn
+            & "./build/Release/stable_diffusion.exe" -m ./models/stable_diffusion_v1_5_ov -t FP16 --dynamic

.gitignore

@@ -0,0 +1,41 @@
+# build/artifact dirs
+_*
+[Bb]uild*/
+cmake-build*
+
+# but ensure we don't skip __init__.py and __main__.py
+!__init__.py
+!__main__.py
+
+# developer tools
+*.idea
+.vscode
+.vs/
+.vsconan/
+.DS_Store
+**/tags
+compile_commands.json
+bin/
+.local_vimrc
+.gdb_history
+.vimspector.json
+doc/
+docs/build_documentation/work_dir/
+temp/
+.repo/
+CMakeLists.txt.user
+docs/IE_PLUGIN_DG/html/
+CMakeUserPresets.json
+
+*.project
+*.cproject
+*.pydevproject
+*.settings
+*/gen/
+*.swp
+/config.xml
+
+# Python-specific
+*.?env*
+*.pyc
+__pycache__

image_generation/common/diffusers/include/scheduler_lcm.hpp

@@ -19,7 +19,8 @@ class LCMScheduler : public Scheduler {
                  float clip_sample_range = 1.0f,
                  float dynamic_thresholding_ratio = 0.995f,
                  float sample_max_value = 1.0f,
-                 bool read_torch_noise = false);
+                 bool read_torch_noise = false,
+                 uint32_t seed = 42);
 
     void set_timesteps(size_t num_inference_steps) override;
 
@@ -47,6 +48,10 @@ class LCMScheduler : public Scheduler {
     float dynamic_thresholding_ratio;
     float sample_max_value;
     bool read_torch_noise;
+    std::mt19937 gen;
+    std::normal_distribution<float> normal;
+    uint32_t seed;
 
     std::vector<float> threshold_sample(const std::vector<float>& flat_sample);
+    std::vector<float> randn_function(uint32_t size, uint32_t seed);
 };

image_generation/common/diffusers/src/scheduler_lcm.cpp

@@ -30,19 +30,6 @@ std::vector<T> linspace(U start, U end, size_t num, bool endpoint = false) {
     return indices;
 }
 
-// for torch.randn()
-std::vector<float> randn_function(uint32_t size, uint32_t seed = 42) {
-    std::vector<float> noise(size);
-    {
-        std::mt19937 gen{static_cast<unsigned long>(seed)};
-        std::normal_distribution<float> normal{0.0f, 1.0f};
-        std::for_each(noise.begin(), noise.end(), [&](float& x) {
-            x = normal(gen);
-        });
-    }
-    return noise;
-}
-
 std::vector<float> read_vector_from_txt(std::string& file_name) {
     std::ifstream input_data(file_name, std::ifstream::in);
     std::istream_iterator<float> start(input_data), end;
@@ -64,7 +51,8 @@ LCMScheduler::LCMScheduler(size_t num_train_timesteps,
                            float clip_sample_range,
                            float dynamic_thresholding_ratio,
                            float sample_max_value,
-                           bool read_torch_noise):
+                           bool read_torch_noise,
+                           uint32_t seed):
                            prediction_type_config(prediction_type),
                            num_train_timesteps_config(num_train_timesteps),
                            original_inference_steps_config(original_inference_steps),
@@ -74,7 +62,10 @@ LCMScheduler::LCMScheduler(size_t num_train_timesteps,
                            clip_sample_range(clip_sample_range),
                            dynamic_thresholding_ratio(dynamic_thresholding_ratio),
                            sample_max_value(sample_max_value),
-                           read_torch_noise(read_torch_noise) {
+                           read_torch_noise(read_torch_noise),
+                           seed(seed),
+                           gen(seed),
+                           normal(0.0f, 1.0f) {
 
     sigma_data = 0.5f; // Default: 0.5
 
@@ -204,7 +195,7 @@ std::map<std::string, ov::Tensor> LCMScheduler::step(ov::Tensor noise_pred, ov::
             std::string noise_file = "../scripts/torch_noise_step_" + std::to_string(inference_step) + ".txt";
             noise = read_vector_from_txt(noise_file);
         } else {
-            noise = randn_function(noise_pred.get_size());
+            noise = randn_function(noise_pred.get_size(), seed);
         }        
 
         for (std::size_t i = 0; i < noise_pred.get_size(); ++i) {
@@ -261,3 +252,13 @@ std::vector<float> LCMScheduler::threshold_sample(const std::vector<float>& flat
 
     return thresholded_sample;
 }
+
+std::vector<float> LCMScheduler::randn_function(uint32_t size, uint32_t seed = 42) {
+    std::vector<float> noise(size);
+    {
+        std::for_each(noise.begin(), noise.end(), [&](float& x) {
+            x = normal(gen);
+        });
+    }
+    return noise;
+}

image_generation/lcm_dreamshaper_v7/cpp/README.md

@@ -24,6 +24,7 @@ conda install -c conda-forge openvino c-compiler cxx-compiler make
 1. Install dependencies to import models from HuggingFace:
 
     ```shell
+    git submodule update --init
     conda activate openvino_lcm_cpp
     python -m pip install -r scripts/requirements.txt
     python -m pip install ../../../thirdparty/openvino_tokenizers/[transformers]

image_generation/lcm_dreamshaper_v7/cpp/scripts/requirements.txt

@@ -1,4 +1,4 @@
 --extra-index-url https://download.pytorch.org/whl/cpu
-torch==2.2.0+cpu
-diffusers==0.26.3
-optimum-intel[nncf,openvino]==1.15.2
+torch==2.2.2+cpu
+diffusers==0.27.2
+optimum-intel[nncf,openvino]==1.16.0

image_generation/lcm_dreamshaper_v7/cpp/src/main.cpp

@@ -264,7 +264,7 @@ int32_t main(int32_t argc, char* argv[]) try {
     std::shared_ptr<Scheduler> scheduler = std::make_shared<LCMScheduler>(LCMScheduler(
         1000, 0.00085f, 0.012f, BetaSchedule::SCALED_LINEAR,
         PredictionType::EPSILON, {}, 50, true, 10.0f, false,
-        false, 1.0f, 0.995f, 1.0f, read_np_latent));
+        false, 1.0f, 0.995f, 1.0f, read_np_latent, user_seed));
     scheduler->set_timesteps(num_inference_steps);
     std::vector<std::int64_t> timesteps = scheduler->get_timesteps();
 

image_generation/stable_diffusion_1_5/cpp/.gitignore

@@ -0,0 +1,3 @@
+build
+images
+models