merge from upstream

openvinotoolkit · Jan 3, 2025 · 903673c · 903673c
2 parents 9a1863b + 42f3053
commit 903673c
Show file tree

Hide file tree

Showing 152 changed files with 6,238 additions and 4,396 deletions.
diff --git a/.github/labeler.yml b/.github/labeler.yml
@@ -13,17 +13,20 @@
 - 'src/python/py_tokenizer.cpp'
 - 'thirdparty/openvino_tokenizers'
 - 'tests/python_tests/tokenizer_configs.py'
+- 'tests/python_tests/test_tokenizer.py'
 
 'category: LLM':
 - 'src/cpp/include/openvino/genai/llm_pipeline.hpp'
 - 'src/cpp/src/llm_pipeline.cpp'
+- 'src/cpp/src/lm_encoding.hpp'
 - 'src/cpp/src/lm_encoding.cpp'
 - 'src/cpp/src/llm_pipeline_base.hpp'
 - 'src/cpp/src/llm_pipeline_static.hpp'
 - 'src/cpp/src/llm_pipeline_static.cpp'
+- 'src/cpp/src/text_callback_streamer.cpp'
+- 'src/cpp/src/text_callback_streamer.hpp'
 - 'src/python/py_llm_pipeline.cpp'
-- 'tests/python_tests/test_generate_api.py'
-- 'tests/python_tests/test_chat_generate_api.py'
+- 'tests/python_tests/test_llm_pipeline.py'
 
 'category: sampling':
 - 'src/cpp/include/openvino/genai/generation_config.hpp'
@@ -35,6 +38,7 @@
 - 'tests/cpp/logit_filtering.cpp'
 - 'tests/cpp/generate_config.cpp'
 - 'tests/cpp/sampler.cpp'
+- 'tests/python_tests/test_sampling.py'
 
 'category: LoRA':
 - 'src/cpp/include/openvino/genai/lora_adapter.hpp'
@@ -54,9 +58,12 @@
 - 'src/cpp/include/openvino/genai/whisper_pipeline.hpp'
 - 'src/cpp/src/whisper/**/*'
 - 'src/cpp/src/whisper_generation_config.cpp'
+- 'src/cpp/src/whisper_pipeline_base.hpp'
 - 'src/cpp/src/whisper_pipeline.cpp'
+- 'src/cpp/src/whisper_pipeline_static.cpp'
+- 'src/cpp/src/whisper_pipeline_static.hpp'
 - 'src/python/py_whisper_pipeline.cpp'
-- 'tests/python_tests/test_whisper_generate_api.py'
+- 'tests/python_tests/test_whisper_pipeline.py'
 
 'category: Python API':
 - 'src/python/**/*'
@@ -65,10 +72,14 @@
 - 'src/include/openvino/genai/visual_language/**/*'
 - 'src/cpp/src/visual_language/**/*'
 - 'src/python/py_vlm_pipeline.cpp'
-- 'tests/python_tests/test_vlm_api.py'
+- 'tests/python_tests/test_vlm_pipeline.py'
 
 'category: speculative decoding':
 - 'src/cpp/src/speculative_decoding/**/*'
+- 'tests/cpp/speculative_decoding.cpp'
+
+'category: prompt lookup':
+- 'src/cpp/src/prompt_lookup/**/*'
 
 'category: continuous batching':
 - 'src/cpp/include/openvino/genai/cache_eviction.hpp'
@@ -91,19 +102,19 @@
 - 'src/cpp/src/generation_handle.cpp'
 - 'src/cpp/src/generation_stream.hpp'
 - 'src/cpp/src/model_runner.hpp'
-- 'src/cpp/src/paged_attention_transformations.cpp'
-- 'src/cpp/src/paged_attention_transformations.hpp'
+- 'src/cpp/src/utils/paged_attention_transformations.cpp'
+- 'src/cpp/src/utils/paged_attention_transformations.hpp'
 - 'src/cpp/src/scheduler.hpp'
 - 'src/cpp/src/sequence_group.cpp'
 - 'src/cpp/src/sequence_group.hpp'
 - 'src/cpp/src/timer.hpp'
 - 'src/python/py_continuous_batching_pipeline.cpp'
-- 'tests/python_tests/test_cache_optimizations.py'
-- 'tests/python_tests/test_preemption.py'
-- 'tests/python_tests/test_sampling.py'
+- 'tests/python_tests/test_continuous_batching.py'
+- 'tests/python_tests/test_kv_cache_eviction.py'
 - 'tests/cpp/block_allocator.cpp'
 - 'tests/cpp/block_hash_store.cpp'
 - 'tests/cpp/block_manager.cpp'
+- 'tests/cpp/cache_eviction.cpp'
 - 'tests/cpp/cache_manager.cpp'
 - 'tests/cpp/device_config.cpp'
 - 'tests/cpp/scheduler.cpp'

diff --git a/.github/workflows/causal_lm_cpp.yml b/.github/workflows/causal_lm_cpp.yml
@@ -16,10 +16,10 @@ concurrency:
   cancel-in-progress: true
 
 env:
-  l_ov_link: https://storage.openvinotoolkit.org/repositories/openvino/packages/nightly/2025.0.0-17539-6abe2e39391/l_openvino_toolkit_ubuntu20_2025.0.0.dev20241205_x86_64.tgz
-  l_u22_ov_link: https://storage.openvinotoolkit.org/repositories/openvino/packages/nightly/2025.0.0-17539-6abe2e39391/l_openvino_toolkit_ubuntu22_2025.0.0.dev20241205_x86_64.tgz
-  m_ov_link: https://storage.openvinotoolkit.org/repositories/openvino/packages/nightly/2025.0.0-17539-6abe2e39391/m_openvino_toolkit_macos_12_6_2025.0.0.dev20241205_x86_64.tgz
-  w_ov_link: https://storage.openvinotoolkit.org/repositories/openvino/packages/nightly/2025.0.0-17539-6abe2e39391/w_openvino_toolkit_windows_2025.0.0.dev20241205_x86_64.zip
+  l_ov_link: https://storage.openvinotoolkit.org/repositories/openvino/packages/nightly/2025.0.0-17726-9ab2c1a18e7/l_openvino_toolkit_ubuntu20_2025.0.0.dev20241230_x86_64.tgz
+  l_u22_ov_link: https://storage.openvinotoolkit.org/repositories/openvino/packages/nightly/2025.0.0-17726-9ab2c1a18e7/l_openvino_toolkit_ubuntu22_2025.0.0.dev20241230_x86_64.tgz
+  m_ov_link: https://storage.openvinotoolkit.org/repositories/openvino/packages/nightly/2025.0.0-17726-9ab2c1a18e7/m_openvino_toolkit_macos_12_6_2025.0.0.dev20241230_x86_64.tgz
+  w_ov_link: https://storage.openvinotoolkit.org/repositories/openvino/packages/nightly/2025.0.0-17726-9ab2c1a18e7/w_openvino_toolkit_windows_2025.0.0.dev20241230_x86_64.zip
 jobs:
   cpp-multinomial-greedy_causal_lm-ubuntu:
     runs-on: ubuntu-20.04-8-cores

diff --git a/.github/workflows/job_vlm_sample_llava.yml b/.github/workflows/job_vlm_sample_llava.yml
@@ -11,7 +11,7 @@ on:
         type: string
 
 env:
-  l_u22_ov_link: https://storage.openvinotoolkit.org/repositories/openvino/packages/nightly/2025.0.0-17539-6abe2e39391/l_openvino_toolkit_ubuntu22_2025.0.0.dev20241205_x86_64.tgz
+  l_u22_ov_link: https://storage.openvinotoolkit.org/repositories/openvino/packages/nightly/2025.0.0-17726-9ab2c1a18e7/l_openvino_toolkit_ubuntu22_2025.0.0.dev20241230_x86_64.tgz
 
 jobs:
   visual_language_chat_sample-ubuntu-llava:

diff --git a/.github/workflows/lcm_dreamshaper_cpp.yml b/.github/workflows/lcm_dreamshaper_cpp.yml
@@ -18,8 +18,8 @@ concurrency:
 
 env:
   PYTHON_VERSION: '3.9'
-  LINUX_OV_ARCHIVE_URL: https://storage.openvinotoolkit.org/repositories/openvino/packages/nightly/2025.0.0-17539-6abe2e39391/l_openvino_toolkit_ubuntu22_2025.0.0.dev20241205_x86_64.tgz
-  WINDOWS_OV_ARCHIVE_URL: https://storage.openvinotoolkit.org/repositories/openvino/packages/nightly/2025.0.0-17539-6abe2e39391/w_openvino_toolkit_windows_2025.0.0.dev20241205_x86_64.zip
+  LINUX_OV_ARCHIVE_URL: https://storage.openvinotoolkit.org/repositories/openvino/packages/nightly/2025.0.0-17726-9ab2c1a18e7/l_openvino_toolkit_ubuntu22_2025.0.0.dev20241230_x86_64.tgz
+  WINDOWS_OV_ARCHIVE_URL: https://storage.openvinotoolkit.org/repositories/openvino/packages/nightly/2025.0.0-17726-9ab2c1a18e7/w_openvino_toolkit_windows_2025.0.0.dev20241230_x86_64.zip
   OV_INSTALL_DIR: ${{ github.workspace }}/ov
 
 jobs:

diff --git a/.github/workflows/linux.yml b/.github/workflows/linux.yml
@@ -109,10 +109,10 @@ jobs:
           merge-multiple: true
 
       - name: CMake Build
-        run: |    
+        run: |
           source ${{ env.OV_INSTALL_DIR }}/setupvars.sh
           cmake -DCMAKE_BUILD_TYPE=${{ matrix.build-type }} -S ${{ env.SRC_DIR}} -B ${{ env.BUILD_DIR }}
-          cmake --build ${{ env.BUILD_DIR}} --config ${{ matrix.build-type }} --parallel $(nproc)
+          cmake --build ${{ env.BUILD_DIR}} --config ${{ matrix.build-type }} --parallel $(nproc) --verbose
           cmake --install ${{ env.BUILD_DIR }} --config ${{ matrix.build-type }} --prefix ${{ env.INSTALL_DIR }}
       
       - name: Pack Artifacts
@@ -268,9 +268,9 @@ jobs:
       matrix:
         test:
           - name: 'Whisper'
-            cmd: 'tests/python_tests/test_whisper_generate_api.py'
+            cmd: 'tests/python_tests/test_whisper_pipeline.py'
           - name: 'LLM & VLM'
-            cmd: 'tests/python_tests --ignore tests/python_tests/test_whisper_generate_api.py -k "not Qwen2-0.5B-Instruct"' # Skip failed tests Qwen2-0.5B-Instruct
+            cmd: 'tests/python_tests --ignore tests/python_tests/test_whisper_pipeline.py'
     defaults:
       run:
         shell: bash

diff --git a/.github/workflows/llm_bench-python.yml b/.github/workflows/llm_bench-python.yml
@@ -114,14 +114,14 @@ jobs:
       - name: Test OpenVINO/LCM_Dreamshaper_v7-int8-ov on Linux Optimum Intel
         run: |
           huggingface-cli download OpenVINO/LCM_Dreamshaper_v7-int8-ov --local-dir ov_models/lcm_dreamshaper_v7
-          python ./tools/llm_bench/benchmark.py -m ./ov_models/lcm_dreamshaper_v7/ -pf ./tools/llm_bench/prompts/stable-diffusion.jsonl -d cpu -n 1 --optimum -ic 4
+          python ./tools/llm_bench/benchmark.py -m ./ov_models/lcm_dreamshaper_v7/ -pf ./tools/llm_bench/prompts/stable-diffusion.jsonl -d cpu -n 1 --optimum --num_steps 4
       - name: Test OpenVINO/LCM_Dreamshaper_v7-int8-ov on Linux with GenAI
         run: |
-          python ./tools/llm_bench/benchmark.py -m ./ov_models/lcm_dreamshaper_v7/ -pf ./tools/llm_bench/prompts/stable-diffusion.jsonl -d cpu -n 1 -ic 4
+          python ./tools/llm_bench/benchmark.py -m ./ov_models/lcm_dreamshaper_v7/ -pf ./tools/llm_bench/prompts/stable-diffusion.jsonl -d cpu -n 1 --num_steps 4
       - name: Test OpenVINO/LCM_Dreamshaper_v7-int8-ov on Linux with GenAI and LoRA
         run: |
           wget -O ./ov_models/soulcard.safetensors https://civitai.com/api/download/models/72591
-          python ./tools/llm_bench/benchmark.py -m ./ov_models/lcm_dreamshaper_v7/ -pf ./tools/llm_bench/prompts/stable-diffusion.jsonl -d cpu -n 1 --lora ./ov_models/soulcard.safetensors --lora_alphas 0.7 -ic 4
+          python ./tools/llm_bench/benchmark.py -m ./ov_models/lcm_dreamshaper_v7/ -pf ./tools/llm_bench/prompts/stable-diffusion.jsonl -d cpu -n 1 --lora ./ov_models/soulcard.safetensors --lora_alphas 0.7 --num_steps 4
           rm -rf ./ov_models/lcm_dreamshaper_v7/
       - name: Test TinyLlama-1.1B-Chat-v1.0 in Speculative Deconding mode on Linux
         run: |

diff --git a/.github/workflows/mac.yml b/.github/workflows/mac.yml
@@ -17,7 +17,7 @@ concurrency:
 
 env:
   PYTHON_VERSION: '3.9'
-  OV_BRANCH: 0080d90974ca84f9a6d359da3388a2a18a93b753
+  OV_BRANCH: master
   OV_TARBALL: ''
 
 jobs:
@@ -178,7 +178,7 @@ jobs:
     if: |
       always() &&
       (needs.openvino_download.outputs.status == 'success' || needs.openvino_build.result == 'success')
-    timeout-minutes: 90
+    timeout-minutes: 120
     defaults:
       run:
         shell: bash
@@ -219,13 +219,13 @@ jobs:
         run: |
           source ${OV_INSTALL_DIR}/setupvars.sh
           cmake -DCMAKE_BUILD_TYPE=Release -S ./ -B ./build/
-          cmake --build ./build/ --config Release -j
+          cmake --build ./build/ --config Release --parallel --verbose
 
       - name: Test bindings
         run: |
           source ${OV_INSTALL_DIR}/setupvars.sh
           python -m pip install ./thirdparty/openvino_tokenizers/[transformers] -r ./tests/python_tests/requirements.txt --find-links ${OV_INSTALL_DIR}/wheels
-          python -m pytest -v ./tests/python_tests/test_chat_generate_api.py::test_set_chat_template
+          python -m pytest -v ./tests/python_tests/test_tokenizer.py::test_set_chat_template
         env:
           PYTHONPATH: "./build/:$PYTHONPATH"
 
@@ -235,7 +235,7 @@ jobs:
           python -m pip install . --verbose --find-links ${OV_INSTALL_DIR}/wheels
           python -c "from openvino_genai import LLMPipeline"
           python -m pip install ./tools/who_what_benchmark --find-links ${OV_INSTALL_DIR}/wheels
-          python -m pytest -v ./tests/python_tests/ --ignore ./tests/python_tests/test_whisper_generate_api.py --ignore ./tests/python_tests/test_vlm_api.py -k "not test_set_chat_template"
+          python -m pytest -v ./tests/python_tests/ --ignore ./tests/python_tests/test_whisper_pipeline.py --ignore ./tests/python_tests/test_vlm_pipeline.py -k "not test_set_chat_template"
 
   genai_python_lib_whisper:
     name: OpenVINO genai extension whisper tests (cmake + wheel)
@@ -284,13 +284,13 @@ jobs:
         run: |
           source ${OV_INSTALL_DIR}/setupvars.sh
           cmake -DCMAKE_BUILD_TYPE=Release -S ./ -B ./build/
-          cmake --build ./build/ --config Release --target py_openvino_genai -j
+          cmake --build ./build/ --config Release --target py_openvino_genai --parallel --verbose
 
       - name: Test bindings
         run: |
           source ${OV_INSTALL_DIR}/setupvars.sh
           python -m pip install ./thirdparty/openvino_tokenizers/[transformers] -r ./tests/python_tests/requirements.txt --find-links ${OV_INSTALL_DIR}/wheels
-          python -m pytest -v ./tests/python_tests/test_whisper_generate_api.py -k test_smoke
+          python -m pytest -v ./tests/python_tests/test_whisper_pipeline.py -k test_smoke
         env:
           PYTHONPATH: "./build/:$PYTHONPATH"
 
@@ -300,7 +300,7 @@ jobs:
           python -m pip install . --verbose --find-links ${OV_INSTALL_DIR}/wheels
           python -c "from openvino_genai import LLMPipeline"
           python -m pip install ./tools/who_what_benchmark --find-links ${OV_INSTALL_DIR}/wheels
-          python -m pytest -v ./tests/python_tests/test_whisper_generate_api.py -k "not test_smoke"
+          python -m pytest -v ./tests/python_tests/test_whisper_pipeline.py -k "not test_smoke"
 
   genai_package:
     name: OpenVINO genai extension (install to OpenVINO package)
@@ -350,7 +350,7 @@ jobs:
         run: |
           source ${OV_INSTALL_DIR}/setupvars.sh
           cmake -DCMAKE_BUILD_TYPE=${{ matrix.build-type }} -S ./ -B ./build/
-          cmake --build ./build/ --config ${{ matrix.build-type }} --target package -j
+          cmake --build ./build/ --config ${{ matrix.build-type }} --target package --parallel --verbose
 
       - name: Build and Install dependencies
         run: |

diff --git a/.github/workflows/stable_diffusion_1_5_cpp.yml b/.github/workflows/stable_diffusion_1_5_cpp.yml
@@ -122,6 +122,8 @@ jobs:
           source openvino_sd_cpp/bin/activate
           optimum-cli export openvino --model dreamlike-art/dreamlike-anime-1.0 --weight-format fp16 --task stable-diffusion models/dreamlike-art-dreamlike-anime-1.0/FP16
           wget -O ./models/soulcard.safetensors https://civitai.com/api/download/models/72591
+        env:
+          HF_HUB_ENABLE_HF_TRANSFER: 1
 
       - name: Run text2image app
         run: |
@@ -198,6 +200,8 @@ jobs:
           . "./openvino_sd_cpp/Scripts/Activate.ps1"
           optimum-cli export openvino --model dreamlike-art/dreamlike-anime-1.0 --task stable-diffusion --weight-format fp16 models/dreamlike-art-dreamlike-anime-1.0/FP16
           Invoke-WebRequest -Uri 'https://civitai.com/api/download/models/72591' -OutFile 'models/soulcard.safetensors'
+        env:
+          HF_HUB_ENABLE_HF_TRANSFER: 1
 
       - name: Run text2image app
         run: |

diff --git a/.github/workflows/windows.yml b/.github/workflows/windows.yml
@@ -17,7 +17,7 @@ concurrency:
 
 env:
   PYTHON_VERSION: '3.11'
-  OV_BRANCH: 0080d90974ca84f9a6d359da3388a2a18a93b753
+  OV_BRANCH: master
   OV_TARBALL: ''
 
 jobs:
@@ -230,13 +230,13 @@ jobs:
         run: |
           . "${{ env.OV_INSTALL_DIR }}/setupvars.ps1"
           cmake -DCMAKE_BUILD_TYPE=Release -S ./ -B ./build/
-          cmake --build ./build/ --config Release -j
+          cmake --build ./build/ --config Release --parallel --verbose
 
       - name: Test bindings
         run: |
           . "${{ env.OV_INSTALL_DIR }}/setupvars.ps1"
           python -m pip install ./thirdparty/openvino_tokenizers/[transformers] -r ./tests/python_tests/requirements.txt --find-links ${env:OV_INSTALL_DIR}/wheels
-          python -m pytest -v ./tests/python_tests/test_chat_generate_api.py::test_set_chat_template
+          python -m pytest -v ./tests/python_tests/test_tokenizer.py::test_set_chat_template
         env:
           PYTHONPATH: "./build/" # cmd evaluates variables in a different way. Setting PYTHONPATH before setupvars.bat instead of doing that after solves that.
 
@@ -245,7 +245,7 @@ jobs:
           . "${{ env.OV_INSTALL_DIR }}/setupvars.ps1"
           python -m pip install . --verbose --find-links ${env:OV_INSTALL_DIR}/wheels
           python -m pip install ./tools/who_what_benchmark --find-links ${env:OV_INSTALL_DIR}/wheels
-          python -m pytest -v ./tests/python_tests/ --ignore ./tests/python_tests/test_whisper_generate_api.py --ignore ./tests/python_tests/test_vlm_api.py -k "not test_set_chat_template"
+          python -m pytest -v ./tests/python_tests/ --ignore ./tests/python_tests/test_whisper_pipeline.py --ignore ./tests/python_tests/test_vlm_pipeline.py -k "not test_set_chat_template"
 
   genai_python_lib_whisper:
     name: OpenVINO genai extension whisper tests (cmake + wheel)
@@ -295,13 +295,13 @@ jobs:
         run: |
           . "${{ env.OV_INSTALL_DIR }}/setupvars.ps1"
           cmake -DCMAKE_BUILD_TYPE=Release -S ./ -B ./build/
-          cmake --build ./build/ --config Release --target py_openvino_genai -j
+          cmake --build ./build/ --config Release --target py_openvino_genai --parallel --verbose
 
       - name: Test bindings
         run: |
           . "${{ env.OV_INSTALL_DIR }}/setupvars.ps1"
           python -m pip install ./thirdparty/openvino_tokenizers/[transformers] -r ./tests/python_tests/requirements.txt --find-links ${env:OV_INSTALL_DIR}/wheels
-          python -m pytest -v ./tests/python_tests/test_whisper_generate_api.py -k test_smoke
+          python -m pytest -v ./tests/python_tests/test_whisper_pipeline.py -k test_smoke
         env:
           PYTHONPATH: "./build/" # cmd evaluates variables in a different way. Setting PYTHONPATH before setupvars.bat instead of doing that after solves that.
 
@@ -310,7 +310,7 @@ jobs:
           . "${{ env.OV_INSTALL_DIR }}/setupvars.ps1"
           python -m pip install . --verbose --find-links ${env:OV_INSTALL_DIR}/wheels
           python -m pip install ./tools/who_what_benchmark --find-links ${env:OV_INSTALL_DIR}/wheels
-          python -m pytest -v ./tests/python_tests/test_whisper_generate_api.py -k "not test_smoke"
+          python -m pytest -v ./tests/python_tests/test_whisper_pipeline.py -k "not test_smoke"
 
   genai_python_lib_vlm:
     name: OpenVINO genai VLM tests (cmake + wheel)
@@ -360,13 +360,13 @@ jobs:
         run: |
           . "${{ env.OV_INSTALL_DIR }}/setupvars.ps1"
           cmake -DCMAKE_BUILD_TYPE=Release -S ./ -B ./build/
-          cmake --build ./build/ --config Release --target py_openvino_genai -j
+          cmake --build ./build/ --config Release --target py_openvino_genai --parallel --verbose
 
       - name: Test bindings
         run: |
           . "${{ env.OV_INSTALL_DIR }}/setupvars.ps1"
           python -m pip install ./thirdparty/openvino_tokenizers/[transformers] -r ./tests/python_tests/requirements.txt --find-links ${env:OV_INSTALL_DIR}/wheels
-          python -m pytest -v ./tests/python_tests/test_vlm_api.py
+          python -m pytest -v ./tests/python_tests/test_vlm_pipeline.py
         env:
           PYTHONPATH: "./build/" # cmd evaluates variables in a different way. Setting PYTHONPATH before setupvars.bat instead of doing that after solves that.
 

diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -85,7 +85,6 @@ if(MSVC AND MSVC_VERSION GREATER_EQUAL 1930 AND MSVC_VERSION LESS 1941)
     add_compile_definitions(_DISABLE_CONSTEXPR_MUTEX_CONSTRUCTOR)
 endif()
 
-
 add_subdirectory(thirdparty)
 add_subdirectory(src)
 if(EXISTS "${OpenVINOGenAI_SOURCE_DIR}/samples")

diff --git a/README.md b/README.md
@@ -331,10 +331,14 @@ For more examples check out our [Generative AI workflow](https://docs.openvino.a
 
 NOTE: Whisper Pipeline requires preprocessing of audio input (to adjust sampling rate and normalize)
  
- ### Converting and compressing image generation model from Hugging Face library
+ ### Converting and quantizing speech-to-text model from Hugging Face library
 ```sh
 #Download and convert to OpenVINO whisper-base model
 optimum-cli export openvino --trust-remote-code --model openai/whisper-base whisper-base
+
+#Download, convert and apply int8 static quantization to whisper-base model
+optimum-cli export openvino --trust-remote-code --model openai/whisper-base \
+--quant-mode int8 --dataset librispeech --num-samples 32 whisper-base-int8
 ```
 
 ### Run generation using Whisper Pipeline API in Python

diff --git a/pyproject.toml b/pyproject.toml
@@ -51,7 +51,7 @@ options = {"BUILD_TOKENIZERS" = "OFF"}
 
 [build-system]
 requires = [
-    "py-build-cmake==0.3.3",
+    "py-build-cmake==0.3.4",
     "openvino~=2025.0.0.0.dev",
     "pybind11-stubgen==2.5.1",
     "cmake~=3.23.0"

diff --git a/samples/cpp/beam_search_causal_lm/beam_search_causal_lm.cpp b/samples/cpp/beam_search_causal_lm/beam_search_causal_lm.cpp
@@ -17,6 +17,7 @@ int main(int argc, char* argv[]) try {
     config.max_new_tokens = 20;
     config.num_beam_groups = 3;
     config.num_beams = 15;
+    config.diversity_penalty = 1.0f;
     config.num_return_sequences = config.num_beams;
 
     // Since the streamer is set, the results will