Merge pull request #331 from yatarkan/yt/sd-to-optimum-cli

Move from image_generation python conversion scripts to optimum-cli

.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

image_generation/stable_diffusion_1_5/cpp/.gitignore

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

image_generation/stable_diffusion_1_5/cpp/README.md

@@ -6,6 +6,10 @@ The pure C++ text-to-image pipeline, driven by the OpenVINO native C++ API for S
 
 ## Step 1: Prepare build environment
 
+Prerequisites:
+- Conda ([installation guide](https://conda.io/projects/conda/en/latest/user-guide/install/index.html))
+
+
 C++ Packages:
 * [CMake](https://cmake.org/download/): Cross-platform build tool
 * [OpenVINO](https://docs.openvino.ai/install): Model inference
@@ -14,7 +18,9 @@ Prepare a python environment and install dependencies:
 ```shell
 conda create -n openvino_sd_cpp python==3.10
 conda activate openvino_sd_cpp
-conda install openvino c-compiler cxx-compiler make
+conda install -c conda-forge openvino c-compiler cxx-compiler make cmake
+# Ensure that Conda standard libraries are used
+conda env config vars set LD_LIBRARY_PATH=$CONDA_PREFIX/lib:$LD_LIBRARY_PATH
 ```
 
 ## Step 2: Convert Stable Diffusion v1.5 and Tokenizer models
@@ -24,32 +30,30 @@ conda install openvino c-compiler cxx-compiler make
 1. Install dependencies to import models from HuggingFace:
 ```shell
 git submodule update --init
+# Reactivate Conda environment after installing dependencies and setting env vars
 conda activate openvino_sd_cpp
-python -m pip install -r scripts/requirements.txt
+python -m pip install -r requirements.txt
 python -m pip install ../../../thirdparty/openvino_tokenizers/[transformers]
 ```
 2. Download a huggingface SD v1.5 model like:
 - [runwayml/stable-diffusion-v1-5](https://huggingface.co/runwayml/stable-diffusion-v1-5)
 - [dreamlike-anime-1.0](https://huggingface.co/dreamlike-art/dreamlike-anime-1.0) to run Stable Diffusion with LoRA adapters.
 
+   Example command for downloading and exporting FP16 model:
+   ```shell
+   export MODEL_PATH="models/dreamlike_anime_1_0_ov/FP16"
+   # Using optimum-cli for exporting model to OpenVINO format
+   optimum-cli export openvino --model dreamlike-art/dreamlike-anime-1.0 --task stable-diffusion --convert-tokenizer --weight-format fp16 $MODEL_PATH
+   # Converting tokenizer manually (`--convert-tokenizer` flag of `optimum-cli` results in "OpenVINO Tokenizer export for CLIPTokenizer is not supported.")
+   convert_tokenizer $MODEL_PATH/tokenizer/ --tokenizer-output-type i32 -o $MODEL_PATH/tokenizer/
+   ```
 
-Example command:
-```shell
-huggingface-cli download --resume-download --local-dir-use-symlinks False dreamlike-art/dreamlike-anime-1.0 --local-dir models/dreamlike-anime-1.0
-```
+   You can also choose other precision and export FP32 or INT8 model.
 
-Please, refer to the official website for [model downloading](https://huggingface.co/docs/hub/models-downloading) to read more details.
-
-3. Run model conversion script to convert PyTorch model to OpenVINO IR via [optimum-intel](https://github.com/huggingface/optimum-intel). Please, use the script `scripts/convert_model.py` to convert the model into `FP16_static` or `FP16_dyn`, which will be saved into the `models` folder:
-```shell
-cd scripts
-python convert_model.py -b 1 -t FP16 -sd ../models/dreamlike-anime-1.0 # to convert to models with static shapes
-python convert_model.py -b 1 -t FP16 -sd ../models/dreamlike-anime-1.0 -dyn True # to keep models with dynamic shapes
-python convert_model.py -b 1 -t INT8 -sd ../models/dreamlike-anime-1.0 -dyn True # to compress the models to INT8
-```
+   Please, refer to the official website for [🤗 Optimum](https://huggingface.co/docs/optimum/main/en/index) and [optimum-intel](https://github.com/huggingface/optimum-intel) to read more details.
 
 > [!NOTE]
->Now the pipeline support batch size = 1 only, i.e. static model `(1, 3, 512, 512)`
+> Now the pipeline support batch size = 1 only, i.e. static model `(1, 3, 512, 512)`
 
 ### LoRA enabling with safetensors
 
@@ -70,7 +74,7 @@ cmake --build build --parallel
 
 ## Step 4: Run Pipeline
 ```shell
-./stable_diffusion [-p <posPrompt>] [-n <negPrompt>] [-s <seed>] [--height <output image>] [--width <output image>] [-d <device>] [-r <readNPLatent>] [-l <lora.safetensors>] [-a <alpha>] [-h <help>] [-m <modelPath>] [-t <modelType>]
+./build/stable_diffusion [-p <posPrompt>] [-n <negPrompt>] [-s <seed>] [--height <output image>] [--width <output image>] [-d <device>] [-r <readNPLatent>] [-l <lora.safetensors>] [-a <alpha>] [-h <help>] [-m <modelPath>] [-t <modelType>] [--dynamic]
 
 Usage:
   stable_diffusion [OPTION...]
@@ -86,8 +90,9 @@ Usage:
 * `--width arg`         Width of output image (default: 512)
 * `-c, --useCache`      Use model caching
 * `-r, --readNPLatent`  Read numpy generated latents from file
-* `-m, --modelPath arg` Specify path of SD model IR (default: ../models/dreamlike-anime-1.0)
-* `-t, --type arg`      Specify the type of SD model IR (FP16_static or FP16_dyn) (default: FP16_static)
+* `-m, --modelPath arg` Specify path of SD model IR (default: ../models/dreamlike_anime_1_0_ov)
+* `-t, --type arg`      Specify the type of SD model IRs (FP32, FP16 or INT8) (default: FP16)
+* `--dynamic`           Specify the model input shape to use dynamic shape
 * `-l, --loraPath arg`  Specify path of lora file. (*.safetensors). (default: )
 * `-a, --alpha arg`     alpha for lora (default: 0.75)
 * `-h, --help`          Print usage
@@ -103,15 +108,15 @@ Negative prompt: (empty, here couldn't use OV tokenizer, check the issues for de
 
 Read the numpy latent instead of C++ std lib for the alignment with Python pipeline
 
-* Generate image without lora `./stable_diffusion -r`
+* Generate image without lora `./build/stable_diffusion -r`
 
    ![](./without_lora.bmp)
 
-* Generate image with soulcard lora `./stable_diffusion -r`
+* Generate image with soulcard lora `./build/stable_diffusion -r`
 
    ![](./soulcard_lora.bmp)
 
-* Generate different size image with dynamic model (C++ lib generated latent): `./stable_diffusion -m ../models/dreamlike-anime-1.0 -t FP16_dyn --height 448 --width 704`
+* Generate different size image with dynamic model (C++ lib generated latent): `./build/stable_diffusion -m ./models/dreamlike_anime_1_0_ov -t FP16 --dynamic --height 448 --width 704`
 
    ![](./704x448.bmp)