diff --git a/.github/workflows/genai-tools.yml b/.github/workflows/genai-tools.yml
index 333bee3e11..bd6cb46362 100644
--- a/.github/workflows/genai-tools.yml
+++ b/.github/workflows/genai-tools.yml
@@ -44,7 +44,7 @@ jobs:
       with:
         platform: ubuntu22
         commit_packages_to_provide: wheels
-        revision: latest_available_commit
+        revision: 345163f87953fb0dd8dd590257eb7fc84378da8e
 
   llm_bench:
     name: 'LLM bench tests'
diff --git a/.github/workflows/linux.yml b/.github/workflows/linux.yml
index 0a991e2a54..0d7a5b7bae 100644
--- a/.github/workflows/linux.yml
+++ b/.github/workflows/linux.yml
@@ -52,7 +52,7 @@ jobs:
       with:
         platform: ubuntu22
         commit_packages_to_provide: wheels
-        revision: latest_available_commit
+        revision: 345163f87953fb0dd8dd590257eb7fc84378da8e
 
     - name: Clone docker tag from OpenVINO repo
       uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
diff --git a/.github/workflows/mac.yml b/.github/workflows/mac.yml
index 5402b79e70..062b83fc27 100644
--- a/.github/workflows/mac.yml
+++ b/.github/workflows/mac.yml
@@ -17,7 +17,7 @@ concurrency:
 
 env:
   PYTHON_VERSION: '3.10'
-  OV_BRANCH: master
+  OV_BRANCH: 345163f87953fb0dd8dd590257eb7fc84378da8e
   OV_TARBALL: ''
 
 jobs:
diff --git a/.github/workflows/stable_diffusion_1_5_cpp.yml b/.github/workflows/stable_diffusion_1_5_cpp.yml
index e0bf5371b3..3b01697f26 100644
--- a/.github/workflows/stable_diffusion_1_5_cpp.yml
+++ b/.github/workflows/stable_diffusion_1_5_cpp.yml
@@ -45,7 +45,7 @@ jobs:
       with:
         platform: ubuntu22
         commit_packages_to_provide: wheels
-        revision: latest_available_commit
+        revision: 345163f87953fb0dd8dd590257eb7fc84378da8e
 
   openvino_download_windows:
     name: Download OpenVINO for Windows
@@ -71,7 +71,7 @@ jobs:
       with:
         platform: windows
         commit_packages_to_provide: wheels
-        revision: latest_available_commit
+        revision: 345163f87953fb0dd8dd590257eb7fc84378da8e
 
   stable_diffusion_1_5_cpp-linux:
     runs-on: ubuntu-22.04-8-cores
diff --git a/.github/workflows/windows.yml b/.github/workflows/windows.yml
index e396671b2c..95a713d7a1 100644
--- a/.github/workflows/windows.yml
+++ b/.github/workflows/windows.yml
@@ -17,7 +17,7 @@ concurrency:
 
 env:
   PYTHON_VERSION: '3.11'
-  OV_BRANCH: master
+  OV_BRANCH: 345163f87953fb0dd8dd590257eb7fc84378da8e
   OV_TARBALL: ''
 
 jobs:
diff --git a/CMakeLists.txt b/CMakeLists.txt
index 181132e210..3a67a24bab 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -60,6 +60,7 @@ if(NOT OpenVINODeveloperPackage_FOUND)
 endif()
 
 include(cmake/features.cmake)
+include(cmake/version.cmake)
 
 if(ENABLE_PYTHON)
     # the following two calls are required for cross-compilation
diff --git a/README.md b/README.md
index be3de5e8ce..c5cf799973 100644
--- a/README.md
+++ b/README.md
@@ -133,13 +133,15 @@ from PIL import Image
 
 # Choose GPU instead of CPU in the line below to run the model on Intel integrated or discrete GPU
 pipe = ov_genai.VLMPipeline("./InternVL2-1B", "CPU")
+pipe.start_chat()
 
 image = Image.open("dog.jpg")
 image_data = np.array(image.getdata()).reshape(1, image.size[1], image.size[0], 3).astype(np.uint8)
 image_data = ov.Tensor(image_data)  
 
 prompt = "Can you describe the image?"
-print(pipe.generate(prompt, image=image_data, max_new_tokens=100))
+result = pipe.generate(prompt, image=image_data, max_new_tokens=100)
+print(result.texts[0])
 ```
 
 ### Run generation using VLMPipeline in C++
@@ -392,7 +394,7 @@ See [here](https://openvinotoolkit.github.io/openvino_notebooks/?search=Automati
 
 ## Additional materials
 
-- [List of supported models](https://github.com/openvinotoolkit/openvino.genai/blob/master/src/docs/SUPPORTED_MODELS.md) (NOTE: models can work, but were not tried yet)
+- [List of supported models](https://github.com/openvinotoolkit/openvino.genai/blob/master/SUPPORTED_MODELS.md) (NOTE: models can work, but were not tried yet)
 - [OpenVINO Generative AI workflow](https://docs.openvino.ai/2024/learn-openvino/llm_inference_guide.html)
 - [Optimum-intel and OpenVINO](https://huggingface.co/docs/optimum/intel/openvino/export)
 
diff --git a/src/docs/SUPPORTED_MODELS.md b/SUPPORTED_MODELS.md
similarity index 95%
rename from src/docs/SUPPORTED_MODELS.md
rename to SUPPORTED_MODELS.md
index 44da29ced4..6b45f47890 100644
--- a/src/docs/SUPPORTED_MODELS.md
+++ b/SUPPORTED_MODELS.md
@@ -147,6 +147,8 @@
   </tbody>
 </table>
 
+> [!NOTE]
+> LoRA adapters are supported.
 
 The pipeline can work with other similar topologies produced by `optimum-intel` with the same model signature. The model is required to have the following inputs after the conversion:
 1. `input_ids` contains the tokens.
@@ -165,12 +167,14 @@ The pipeline can work with other similar topologies produced by `optimum-intel`
       <th>Architecture</th>
       <th>Text 2 image</th>
       <th>Image 2 image</th>
+      <th>LoRA support</th>
       <th>Example HuggingFace Models</th>
     </tr>
     <tr>
       <td><code>Latent Consistency Model</code></td>
       <td>Supported</td>
       <td>Supported</td>
+      <td>Supported</td>
       <td>
         <ul>
           <li><a href="https://huggingface.co/SimianLuo/LCM_Dreamshaper_v7"><code>SimianLuo/LCM_Dreamshaper_v7</code></a></li>
@@ -181,6 +185,7 @@ The pipeline can work with other similar topologies produced by `optimum-intel`
       <td><code>Stable Diffusion</code></td>
       <td>Supported</td>
       <td>Supported</td>
+      <td>Supported</td>
       <td>
         <ul>
           <li><a href="https://huggingface.co/CompVis/stable-diffusion-v1-1"><code>CompVis/stable-diffusion-v1-1</code></a></li>
@@ -213,6 +218,7 @@ The pipeline can work with other similar topologies produced by `optimum-intel`
       <td><code>Stable Diffusion XL</code></td>
       <td>Supported</td>
       <td>Supported</td>
+      <td>Supported</td>
       <td>
         <ul>
           <li><a href="https://huggingface.co/stabilityai/stable-diffusion-xl-base-0.9"><code>stabilityai/stable-diffusion-xl-base-0.9</code></a></li>
@@ -225,6 +231,7 @@ The pipeline can work with other similar topologies produced by `optimum-intel`
       <td><code>Stable Diffusion 3</code></td>
       <td>Supported</td>
       <td>Not supported</td>
+      <td>Not supported</td>
       <td>
         <ul>
           <li><a href="https://huggingface.co/stabilityai/stable-diffusion-3-medium-diffusers"><code>stabilityai/stable-diffusion-3-medium-diffusers</code></a></li>
@@ -237,6 +244,7 @@ The pipeline can work with other similar topologies produced by `optimum-intel`
       <td><code>Flux</code></td>
       <td>Supported</td>
       <td>Not supported</td>
+      <td>Not supported</td>
       <td>
         <ul>
           <li><a href="https://huggingface.co/black-forest-labs/FLUX.1-schnell"><code>black-forest-labs/FLUX.1-schnell</code></a></li>
@@ -260,10 +268,12 @@ In addition to image generation models, `InpaintingPipeline` supports specialize
   <tbody style="vertical-align: top;">
     <tr>
       <th>Architecture</th>
+      <th>LoRA support</th>
       <th>Example HuggingFace Models</th>
     </tr>
     <tr>
       <td><code>Stable Diffusion</code></td>
+      <td>Supported</td>
       <td>
         <ul>
           <li><a href="https://huggingface.co/stabilityai/stable-diffusion-2-inpainting"><code>stabilityai/stable-diffusion-2-inpainting</code></a></li>
@@ -275,13 +285,22 @@ In addition to image generation models, `InpaintingPipeline` supports specialize
     </tr>
     <tr>
       <td><code>Stable Diffusion XL</code></td>
+      <td>Supported</td>
       <td>
         <ul>
           <li><a href="https://huggingface.co/diffusers/stable-diffusion-xl-1.0-inpainting-0.1"><code>diffusers/stable-diffusion-xl-1.0-inpainting-0.1</code></a></li>
         </ul>
       </td>
     </tr>
-    </tr>
+    <!-- <tr>
+      <td><code>FLUX</code></td>
+      <td>Not supported</td>
+      <td>
+        <ul>
+          <li><a href="https://huggingface.co/black-forest-labs/FLUX.1-Fill-dev"><code>black-forest-labs/FLUX.1-Fill-dev</code></a></li>
+        </ul>
+      </td>
+    </tr> -->
   </tbody>
 </table>
 
@@ -292,11 +311,13 @@ In addition to image generation models, `InpaintingPipeline` supports specialize
     <tr>
       <th>Architecture</th>
       <th>Models</th>
+      <th>LoRA support</th>
       <th>Example HuggingFace Models</th>
     </tr>
     <tr>
       <td><code>InternVL2</code></td>
       <td>InternVL2</td>
+      <td>Not supported</td>
       <td>
         <ul>
           <li><a href="https://huggingface.co/OpenGVLab/InternVL2-1B"><code>OpenGVLab/InternVL2-1B</code></a></li>
@@ -309,6 +330,7 @@ In addition to image generation models, `InpaintingPipeline` supports specialize
     <tr>
       <td><code>LLaVA</code></td>
       <td>LLaVA-v1.5</td>
+      <td>Not supported</td>
       <td>
         <ul>
           <li><a href="https://huggingface.co/llava-hf/llava-1.5-7b-hf"><code>llava-hf/llava-1.5-7b-hf</code></a></li>
@@ -318,6 +340,7 @@ In addition to image generation models, `InpaintingPipeline` supports specialize
     <tr>
       <td><code>LLaVA-NeXT</code></td>
       <td>LLaVa-v1.6</td>
+      <td>Not supported</td>
       <td>
         <ul>
           <li><a href="https://huggingface.co/llava-hf/llava-v1.6-mistral-7b-hf"><code>llava-hf/llava-v1.6-mistral-7b-hf</code></a></li>
@@ -329,6 +352,7 @@ In addition to image generation models, `InpaintingPipeline` supports specialize
     <tr>
       <td><code>MiniCPMV</code></td>
       <td>MiniCPM-V-2_6</td>
+      <td>Not supported</td>
       <td>
         <ul>
           <li><a href="https://huggingface.co/openbmb/MiniCPM-V-2_6"><code>openbmb/MiniCPM-V-2_6</code></a></li>
@@ -345,11 +369,13 @@ In addition to image generation models, `InpaintingPipeline` supports specialize
     <tr>
       <th>Architecture</th>
       <th>Models</th>
+      <th>LoRA support</th>
       <th>Example HuggingFace Models</th>
     </tr>
     <tr>
       <td rowspan=2><code>WhisperForConditionalGeneration</code></td>
       <td>Whisper</td>
+      <td>Not supported</td>
       <td>
         <ul>
           <li><a href="https://huggingface.co/openai/whisper-tiny"><code>openai/whisper-tiny</code></a></li>
@@ -366,6 +392,7 @@ In addition to image generation models, `InpaintingPipeline` supports specialize
     </tr>
     <tr>
       <td>Distil-Whisper</td>
+      <td>Not supported</td>
       <td>
         <ul>
           <li><a href="https://huggingface.co/distil-whisper/distil-small.en"><code>distil-whisper/distil-small.en</code></a></li>
diff --git a/cmake/templates/__version__.py.in b/cmake/templates/__version__.py.in
deleted file mode 100644
index ce8e01a246..0000000000
--- a/cmake/templates/__version__.py.in
+++ /dev/null
@@ -1,5 +0,0 @@
-# Copyright (C) 2024 Intel Corporation
-# SPDX-License-Identifier: Apache-2.0
-
-# Will be overwritten by cmake.
-__version__ = "@OpenVINOGenAI_VERSION@"
diff --git a/cmake/templates/version.cpp.in b/cmake/templates/version.cpp.in
new file mode 100644
index 0000000000..f6015832f9
--- /dev/null
+++ b/cmake/templates/version.cpp.in
@@ -0,0 +1,19 @@
+// Copyright (C) 2023-2025 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#include "openvino/genai/version.hpp"
+
+namespace ov {
+namespace genai {
+
+const Version get_version() {
+    const static Version version = {
+        "@OpenVINOGenAI_FULL_VERSION@",
+        "OpenVINO GenAI version",
+    };
+
+    return version;
+}
+
+} // namespace genai
+} // namespace ov
diff --git a/cmake/templates/version.hpp.in b/cmake/templates/version.hpp.in
new file mode 100644
index 0000000000..34120ef632
--- /dev/null
+++ b/cmake/templates/version.hpp.in
@@ -0,0 +1,34 @@
+// Copyright (C) 2023-2025 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "openvino/core/version.hpp"
+#include "openvino/genai/visibility.hpp"
+
+/**
+ * OpenVINO GenAI major version
+ */
+#define OPENVINO_GENAI_VERSION_MAJOR @OpenVINOGenAI_VERSION_MAJOR@
+
+/**
+ * OpenVINO GenAI minor version
+ */
+#define OPENVINO_GENAI_VERSION_MINOR @OpenVINOGenAI_VERSION_MINOR@
+
+/**
+ * OpenVINO GenAI patch version
+ */
+#define OPENVINO_GENAI_VERSION_PATCH @OpenVINOGenAI_VERSION_PATCH@
+
+namespace ov {
+namespace genai {
+
+/**
+ * Returns OpenVINO GenAI full version including git commit and hash information in form of:
+ *   <MAJOR>.<MINOR>.<PATCH>.<REVISION>-<COMMIT NUMBER>-<COMMIT HASH>[-<BRANCH SUFFIX>]
+ */
+OPENVINO_EXTERN_C OPENVINO_GENAI_EXPORTS const ov::Version OPENVINO_CDECL get_version();
+
+} // namespace genai
+} // namespace ov
diff --git a/cmake/version.cmake b/cmake/version.cmake
new file mode 100644
index 0000000000..b9b51e8fe2
--- /dev/null
+++ b/cmake/version.cmake
@@ -0,0 +1,72 @@
+# Copyright (C) 2018-2024 Intel Corporation
+# SPDX-License-Identifier: Apache-2.0
+#
+
+find_package(Git QUIET)
+
+function(ov_genai_branch_name VAR)
+    if(GIT_FOUND)
+        execute_process(
+                COMMAND ${GIT_EXECUTABLE} rev-parse --abbrev-ref HEAD
+                WORKING_DIRECTORY ${OpenVINOGenAI_SOURCE_DIR}
+                OUTPUT_VARIABLE GIT_BRANCH
+                RESULT_VARIABLE EXIT_CODE
+                OUTPUT_STRIP_TRAILING_WHITESPACE)
+        if(EXIT_CODE EQUAL 0)
+            set(${VAR} ${GIT_BRANCH} PARENT_SCOPE)
+        endif()
+    endif()
+endfunction()
+
+function(ov_genai_commit_hash VAR)
+    if(GIT_FOUND)
+        execute_process(
+                COMMAND ${GIT_EXECUTABLE} rev-parse --short=11 HEAD
+                WORKING_DIRECTORY ${OpenVINOGenAI_SOURCE_DIR}
+                OUTPUT_VARIABLE GIT_COMMIT_HASH
+                RESULT_VARIABLE EXIT_CODE
+                OUTPUT_STRIP_TRAILING_WHITESPACE)
+        if(EXIT_CODE EQUAL 0)
+            set(${VAR} ${GIT_COMMIT_HASH} PARENT_SCOPE)
+        endif()
+    endif()
+endfunction()
+
+function(ov_genai_commit_number VAR)
+    set(GIT_COMMIT_NUMBER_FOUND OFF)
+    if(GIT_FOUND)
+        execute_process(
+                COMMAND ${GIT_EXECUTABLE} rev-list --count HEAD
+                WORKING_DIRECTORY ${OpenVINOGenAI_SOURCE_DIR}
+                OUTPUT_VARIABLE GIT_COMMIT_NUMBER
+                RESULT_VARIABLE EXIT_CODE
+                OUTPUT_STRIP_TRAILING_WHITESPACE)
+        if(EXIT_CODE EQUAL 0)
+            set(GIT_COMMIT_NUMBER_FOUND ON)
+            set(${VAR} ${GIT_COMMIT_NUMBER} PARENT_SCOPE)
+        endif()
+    endif()
+    if(NOT GIT_COMMIT_NUMBER_FOUND)
+        # set zeros since git is not available
+        set(${VAR} "000" PARENT_SCOPE)
+    endif()
+endfunction()
+
+function(ov_genai_full_version full_version)
+    if(GIT_FOUND)
+        ov_genai_branch_name(GIT_BRANCH)
+        ov_genai_commit_hash(GIT_COMMIT_HASH)
+        ov_genai_commit_number(GIT_COMMIT_NUMBER)
+
+        if(NOT GIT_BRANCH MATCHES "^(master|HEAD)$")
+            set(GIT_BRANCH_POSTFIX "-${GIT_BRANCH}")
+        endif()
+
+        set(${full_version} "${OpenVINOGenAI_VERSION}-${GIT_COMMIT_NUMBER}-${GIT_COMMIT_HASH}${GIT_BRANCH_POSTFIX}" PARENT_SCOPE)
+    else()
+        set(${full_version} "${OpenVINOGenAI_VERSION}" PARENT_SCOPE)
+    endif()
+endfunction()
+
+ov_genai_full_version(OpenVINOGenAI_FULL_VERSION)
+message(STATUS "OpenVINO GenAI full version: ${OpenVINOGenAI_FULL_VERSION}")
diff --git a/llm_bench/python/README.md b/llm_bench/python/README.md
deleted file mode 100644
index 272ed11d1b..0000000000
--- a/llm_bench/python/README.md
+++ /dev/null
@@ -1,4 +0,0 @@
-# Benchmarking Script for Large Language Models
-
-> [!IMPORTANT]  
-> LLM bench code was moved to [tools](../../tools/llm_bench/) directory. Please navigate to the new directory for continue of tool usage.
\ No newline at end of file
diff --git a/llm_bench/python/who_what_benchmark/README.md b/llm_bench/python/who_what_benchmark/README.md
deleted file mode 100644
index 414b4d9342..0000000000
--- a/llm_bench/python/who_what_benchmark/README.md
+++ /dev/null
@@ -1,4 +0,0 @@
-# Simple Accuracy Benchmark for Generative AI models
-
-> [!IMPORTANT]  
-> Who What Benchmark code was moved to [tools](../../../tools/who_what_benchmark/) directory. Please navigate to the new directory for continue of tool usage.
\ No newline at end of file
diff --git a/samples/cpp/visual_language_chat/README.md b/samples/cpp/visual_language_chat/README.md
index 39364d51ee..73baf0088a 100644
--- a/samples/cpp/visual_language_chat/README.md
+++ b/samples/cpp/visual_language_chat/README.md
@@ -29,7 +29,7 @@ Follow [Get Started with Samples](https://docs.openvino.ai/2024/learn-openvino/o
 
 Discrete GPUs (dGPUs) usually provide better performance compared to CPUs. It is recommended to run larger models on a dGPU with 32GB+ RAM. For example, the model `llava-hf/llava-v1.6-mistral-7b-hf` can benefit from being run on a dGPU. Modify the source code to change the device for inference to the `GPU`.
 
-See [SUPPORTED_MODELS.md](../../../src/docs/SUPPORTED_MODELS.md#visual-language-models) for the list of supported models.
+See [SUPPORTED_MODELS.md](../../../SUPPORTED_MODELS.md#visual-language-models) for the list of supported models.
 
 ## Run benchmark:
 
diff --git a/samples/cpp/whisper_speech_recognition/README.md b/samples/cpp/whisper_speech_recognition/README.md
index d649266613..2ea3322dee 100644
--- a/samples/cpp/whisper_speech_recognition/README.md
+++ b/samples/cpp/whisper_speech_recognition/README.md
@@ -31,7 +31,7 @@ Output:
 timestamps: [0, 2] text:  How are you doing today?
 ```
 
-See [SUPPORTED_MODELS.md](../../../src/docs/SUPPORTED_MODELS.md#whisper-models) for the list of supported models.
+See [SUPPORTED_MODELS.md](../../../SUPPORTED_MODELS.md#whisper-models) for the list of supported models.
 
 # Whisper pipeline usage
 
diff --git a/samples/python/multinomial_causal_lm/multinomial_causal_lm.py b/samples/python/multinomial_causal_lm/multinomial_causal_lm.py
index 953388ed6a..5ec9d54601 100755
--- a/samples/python/multinomial_causal_lm/multinomial_causal_lm.py
+++ b/samples/python/multinomial_causal_lm/multinomial_causal_lm.py
@@ -90,7 +90,7 @@ def put(self, token_id: int) -> bool:
             word = text[self.print_len:]            
             self.tokens_cache = []
             self.print_len = 0
-        elif len(text) >= 3 and text[-3:] == chr(65533):
+        elif len(text) >= 3 and text[-1] == chr(65533):
             # Don't print incomplete text.
             pass
         elif len(text) > self.print_len:
diff --git a/samples/python/whisper_speech_recognition/README.md b/samples/python/whisper_speech_recognition/README.md
index aeb46444bf..5f373df2b7 100644
--- a/samples/python/whisper_speech_recognition/README.md
+++ b/samples/python/whisper_speech_recognition/README.md
@@ -38,7 +38,7 @@ Output:
 timestamps: [0, 2] text:  How are you doing today?
 ```
 
-See [SUPPORTED_MODELS.md](../../../src/docs/SUPPORTED_MODELS.md#whisper-models) for the list of supported models.
+See [SUPPORTED_MODELS.md](../../../SUPPORTED_MODELS.md#whisper-models) for the list of supported models.
 
 # Whisper pipeline usage
 
diff --git a/src/cpp/CMakeLists.txt b/src/cpp/CMakeLists.txt
index 24367c17ce..e954037daf 100644
--- a/src/cpp/CMakeLists.txt
+++ b/src/cpp/CMakeLists.txt
@@ -54,9 +54,18 @@ FetchContent_MakeAvailable(safetensors.h)
 
 ov_genai_build_jinja2cpp()
 
+# generate version files
+
+configure_file("${OpenVINOGenAI_SOURCE_DIR}/cmake/templates/version.hpp.in"
+               "${CMAKE_CURRENT_BINARY_DIR}/openvino/genai/version.hpp" @ONLY)
+
+configure_file("${OpenVINOGenAI_SOURCE_DIR}/cmake/templates/version.cpp.in"
+               "${CMAKE_CURRENT_BINARY_DIR}/version.cpp" @ONLY)
+
 # Library
 
 file(GLOB_RECURSE SOURCE_FILES "${CMAKE_CURRENT_SOURCE_DIR}/src/*.cpp" "${CMAKE_CURRENT_SOURCE_DIR}/src/*.c")
+list(APPEND SOURCE_FILES "${CMAKE_CURRENT_BINARY_DIR}/version.cpp")
 
 set(TARGET_NAME openvino_genai)
 
@@ -68,7 +77,9 @@ if(TARGET openvino_tokenizers)
 endif()
 
 target_include_directories(${TARGET_NAME}
-    PUBLIC "$<BUILD_INTERFACE:${CMAKE_CURRENT_SOURCE_DIR}/include>" "$<INSTALL_INTERFACE:runtime/include>"
+    PUBLIC "$<BUILD_INTERFACE:${CMAKE_CURRENT_SOURCE_DIR}/include>"
+           "$<INSTALL_INTERFACE:runtime/include>"
+           "$<BUILD_INTERFACE:${CMAKE_CURRENT_BINARY_DIR}>"
     PRIVATE "${CMAKE_CURRENT_SOURCE_DIR}/src")
 
 target_include_directories(${TARGET_NAME} SYSTEM PRIVATE "${safetensors.h_SOURCE_DIR}")
@@ -101,7 +112,7 @@ endif()
 
 if(OpenVINODeveloperPackage_FOUND)
     # must be called after all target_link_libraries
-    # ov_add_api_validator_post_build_step(TARGET ${TARGET_NAME})
+    ov_add_api_validator_post_build_step(TARGET ${TARGET_NAME})
 
     ov_ncc_naming_style(FOR_TARGET ${TARGET_NAME}
                         SOURCE_DIRECTORIES "${CMAKE_CURRENT_SOURCE_DIR}/include")
@@ -145,6 +156,9 @@ install(TARGETS ${TARGET_NAME} EXPORT OpenVINOGenAITargets
 
 install(DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR}/include/
         DESTINATION runtime/include COMPONENT core_genai_dev)
+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/openvino/genai/version.hpp
+        DESTINATION runtime/include/openvino/genai COMPONENT core_genai_dev)
+
 install(EXPORT OpenVINOGenAITargets FILE OpenVINOGenAITargets.cmake
         NAMESPACE openvino:: DESTINATION runtime/cmake
         COMPONENT core_genai_dev)
diff --git a/src/cpp/src/device_config.hpp b/src/cpp/src/device_config.hpp
index cc2e21b9a1..fee6c7abd1 100644
--- a/src/cpp/src/device_config.hpp
+++ b/src/cpp/src/device_config.hpp
@@ -117,22 +117,22 @@ class DeviceConfig {
         }
 
         for (size_t layer_id = 0; layer_id < m_num_decoder_layers; layer_id++) {
-            m_key_cache_shape.push_back(ov::PartialShape{ov::Dimension::dynamic(),
-                                                         ov::Dimension(m_num_kv_heads[layer_id]),
-                                                         ov::Dimension(m_block_size),
-                                                         ov::Dimension(m_head_size)});
-
             m_value_cache_shape.push_back(ov::PartialShape{ov::Dimension::dynamic(),
                                                            ov::Dimension(m_num_kv_heads[layer_id]),
                                                            ov::Dimension(m_block_size),
                                                            ov::Dimension(m_head_size)});
 
-            if (m_device.find("GPU") != std::string::npos) {
+            if (m_device.find("GPU") == std::string::npos) {
+                m_key_cache_shape.push_back(ov::PartialShape{ov::Dimension::dynamic(),
+                                                             ov::Dimension(m_num_kv_heads[layer_id]),
+                                                             ov::Dimension(m_block_size),
+                                                             ov::Dimension(m_head_size)});
+            } else  if (m_device.find("GPU") != std::string::npos) {
                 // Update key shape, as the key's shape is different from the value's shape
                 m_key_cache_shape.push_back(ov::PartialShape{ov::Dimension::dynamic(),
-                                                     ov::Dimension(m_num_kv_heads[layer_id]),
-                                                     ov::Dimension(m_head_size),
-                                                     ov::Dimension(m_block_size)});
+                                                             ov::Dimension(m_num_kv_heads[layer_id]),
+                                                             ov::Dimension(m_head_size),
+                                                             ov::Dimension(m_block_size)});
             }
         }
     }
diff --git a/src/cpp/src/llm_pipeline_static.cpp b/src/cpp/src/llm_pipeline_static.cpp
index 94aa6e19fe..c98b571179 100644
--- a/src/cpp/src/llm_pipeline_static.cpp
+++ b/src/cpp/src/llm_pipeline_static.cpp
@@ -739,7 +739,10 @@ std::shared_ptr<ov::CompiledModel> StatefulLLMPipeline::setupAndCompileModel(
 
     rename_key(pipeline_config, "PREFILL_CONFIG", "NPUW_LLM_PREFILL_CONFIG");
     rename_key(pipeline_config, "GENERATE_CONFIG", "NPUW_LLM_GENERATE_CONFIG");
-  
+
+    // Replace CACHE_DIR option if NPUW is enabled
+    set_npuw_cache_dir(pipeline_config);
+
     return std::make_shared<ov::CompiledModel>(genai::utils::singleton_core().compile_model(model, "NPU", pipeline_config));
 }
 
diff --git a/src/python/CMakeLists.txt b/src/python/CMakeLists.txt
index 75a2fd59a7..1293246260 100644
--- a/src/python/CMakeLists.txt
+++ b/src/python/CMakeLists.txt
@@ -34,9 +34,6 @@ file(COPY "${CMAKE_CURRENT_SOURCE_DIR}/openvino_genai/__init__.py"
           "${CMAKE_CURRENT_SOURCE_DIR}/openvino_genai/py_openvino_genai.pyi"
      DESTINATION "${CMAKE_BINARY_DIR}/openvino_genai/")
 
-configure_file("${OpenVINOGenAI_SOURCE_DIR}/cmake/templates/__version__.py.in"
-               "${CMAKE_BINARY_DIR}/openvino_genai/__version__.py" @ONLY)
-
 if(OpenVINODeveloperPackage_FOUND)
     # TODO: commit changes separately
     # ov_add_clang_format_target(${TARGET_NAME}_clang FOR_TARGETS ${TARGET_NAME})
@@ -69,18 +66,12 @@ endif()
 install(FILES "${CMAKE_CURRENT_SOURCE_DIR}/openvino_genai/__init__.py"
               "${CMAKE_CURRENT_SOURCE_DIR}/openvino_genai/__init__.pyi"
               "${CMAKE_CURRENT_SOURCE_DIR}/openvino_genai/py_openvino_genai.pyi"
-              "${CMAKE_BINARY_DIR}/openvino_genai/__version__.py"
         DESTINATION python/openvino_genai
         COMPONENT pygenai_${Python3_VERSION_MAJOR}_${Python3_VERSION_MINOR})
 install(TARGETS ${TARGET_NAME}
         LIBRARY DESTINATION python/openvino_genai
         COMPONENT pygenai_${Python3_VERSION_MAJOR}_${Python3_VERSION_MINOR})
 
-install(FILES "${CMAKE_BINARY_DIR}/openvino_genai/__version__.py"
-        DESTINATION openvino_genai
-        COMPONENT wheel_genai
-        EXCLUDE_FROM_ALL)
-
 install(FILES "${OpenVINOGenAI_SOURCE_DIR}/LICENSE"
               "${OpenVINOGenAI_SOURCE_DIR}/third-party-programs.txt"
               "${OpenVINOGenAI_SOURCE_DIR}/SECURITY.md"
@@ -154,7 +145,8 @@ if(pybind11_stubgen_AVAILABLE)
     endif()
 
     set(stub_files_location "${OpenVINOGenAI_BINARY_DIR}/src/python")
-    set(generated_files ${stub_files_location}/openvino_genai/__init__.pyi
+    set(init_pyi_file "${stub_files_location}/openvino_genai/__init__.pyi")
+    set(generated_files ${init_pyi_file}
                         ${stub_files_location}/openvino_genai/py_openvino_genai.pyi)
     set_source_files_properties(${generated_files} PROPERTIES GENERATED ON)
 
@@ -184,6 +176,9 @@ if(pybind11_stubgen_AVAILABLE)
                                             "${CMAKE_BINARY_DIR}/openvino_genai/py_openvino_genai.pyi"
         COMMAND "${CMAKE_COMMAND}" -E env PYTHONPATH=${CMAKE_BINARY_DIR}:${openvino_pythonpath}:$ENV{PYTHONPATH}
                 ${pybind11_stubgen} --output-dir ${stub_files_location} openvino_genai
+        COMMAND "${CMAKE_COMMAND}"
+                -D init_pyi_file=${init_pyi_file}
+                -P "${CMAKE_CURRENT_SOURCE_DIR}/clean_version.cmake"
         ${validation_command}
         ${copy_to_source_command}
         COMMAND "${CMAKE_COMMAND}" -E copy ${generated_files} "${CMAKE_BINARY_DIR}/openvino_genai/"
@@ -192,6 +187,7 @@ if(pybind11_stubgen_AVAILABLE)
             ${python_sources}
             ${validation_dependencies}
             "${CMAKE_CURRENT_SOURCE_DIR}/openvino_genai/__init__.py"
+            "${CMAKE_CURRENT_SOURCE_DIR}/clean_version.cmake"
             "${CMAKE_CURRENT_SOURCE_DIR}/compare_pyi.cmake"
         COMMENT "[${pybind11_stubgen_dep}] Generate .pyi files"
         VERBATIM)
diff --git a/src/python/clean_version.cmake b/src/python/clean_version.cmake
new file mode 100644
index 0000000000..f02e293493
--- /dev/null
+++ b/src/python/clean_version.cmake
@@ -0,0 +1,21 @@
+# Copyright (C) 2024 Intel Corporation
+# SPDX-License-Identifier: Apache-2.0
+#
+
+foreach(var IN ITEMS init_pyi_file)
+    if(NOT DEFINED ${var})
+        message(FATAL_ERROR "Variable ${var} is not defined")
+    endif()
+endforeach()
+
+file(STRINGS ${init_pyi_file} file_lines)
+
+foreach(file_line IN LISTS file_lines)
+    if(file_line MATCHES "^__version__.*")
+        set(file_line "__version__: str")
+    endif()
+
+    set(file_content "${file_content}${file_line}\n")
+endforeach()
+
+file(WRITE ${init_pyi_file} ${file_content})
diff --git a/src/python/openvino_genai/__init__.py b/src/python/openvino_genai/__init__.py
index a0b0faf58c..0ad7ba3f12 100644
--- a/src/python/openvino_genai/__init__.py
+++ b/src/python/openvino_genai/__init__.py
@@ -5,8 +5,6 @@
 
 import openvino  # add_dll_directory for openvino lib
 import os
-from .__version__ import __version__
-
 
 if hasattr(os, "add_dll_directory"):
     os.add_dll_directory(os.path.dirname(__file__))
@@ -17,8 +15,11 @@
     RawPerfMetrics,
     PerfMetrics,
     StreamerBase,
+    get_version,
 )
 
+__version__ = get_version()
+
 # VLM pipeline
 
 from .py_openvino_genai import (
diff --git a/src/python/openvino_genai/__init__.pyi b/src/python/openvino_genai/__init__.pyi
index 187e0a0a06..0a401ae958 100644
--- a/src/python/openvino_genai/__init__.pyi
+++ b/src/python/openvino_genai/__init__.pyi
@@ -42,7 +42,8 @@ from openvino_genai.py_openvino_genai import WhisperPerfMetrics
 from openvino_genai.py_openvino_genai import WhisperPipeline
 from openvino_genai.py_openvino_genai import WhisperRawPerfMetrics
 from openvino_genai.py_openvino_genai import draft_model
+from openvino_genai.py_openvino_genai import get_version
 import os as os
 from . import py_openvino_genai
-__all__ = ['Adapter', 'AdapterConfig', 'AggregationMode', 'AutoencoderKL', 'CLIPTextModel', 'CLIPTextModelWithProjection', 'CacheEvictionConfig', 'ChunkStreamerBase', 'ContinuousBatchingPipeline', 'CppStdGenerator', 'DecodedResults', 'EncodedResults', 'FluxTransformer2DModel', 'GenerationConfig', 'GenerationResult', 'Generator', 'Image2ImagePipeline', 'ImageGenerationConfig', 'InpaintingPipeline', 'LLMPipeline', 'PerfMetrics', 'RawPerfMetrics', 'SD3Transformer2DModel', 'Scheduler', 'SchedulerConfig', 'StopCriteria', 'StreamerBase', 'T5EncoderModel', 'Text2ImagePipeline', 'TokenizedInputs', 'Tokenizer', 'TorchGenerator', 'UNet2DConditionModel', 'VLMPipeline', 'WhisperGenerationConfig', 'WhisperPerfMetrics', 'WhisperPipeline', 'WhisperRawPerfMetrics', 'draft_model', 'openvino', 'os', 'py_openvino_genai']
-__version__: str = '2025.0.0.0'
+__all__ = ['Adapter', 'AdapterConfig', 'AggregationMode', 'AutoencoderKL', 'CLIPTextModel', 'CLIPTextModelWithProjection', 'CacheEvictionConfig', 'ChunkStreamerBase', 'ContinuousBatchingPipeline', 'CppStdGenerator', 'DecodedResults', 'EncodedResults', 'FluxTransformer2DModel', 'GenerationConfig', 'GenerationResult', 'Generator', 'Image2ImagePipeline', 'ImageGenerationConfig', 'InpaintingPipeline', 'LLMPipeline', 'PerfMetrics', 'RawPerfMetrics', 'SD3Transformer2DModel', 'Scheduler', 'SchedulerConfig', 'StopCriteria', 'StreamerBase', 'T5EncoderModel', 'Text2ImagePipeline', 'TokenizedInputs', 'Tokenizer', 'TorchGenerator', 'UNet2DConditionModel', 'VLMPipeline', 'WhisperGenerationConfig', 'WhisperPerfMetrics', 'WhisperPipeline', 'WhisperRawPerfMetrics', 'draft_model', 'get_version', 'openvino', 'os', 'py_openvino_genai']
+__version__: str
diff --git a/src/python/openvino_genai/py_openvino_genai.pyi b/src/python/openvino_genai/py_openvino_genai.pyi
index d405cd9bbf..5adde32db4 100644
--- a/src/python/openvino_genai/py_openvino_genai.pyi
+++ b/src/python/openvino_genai/py_openvino_genai.pyi
@@ -5,7 +5,7 @@ from __future__ import annotations
 import openvino._pyopenvino
 import os
 import typing
-__all__ = ['Adapter', 'AdapterConfig', 'AggregationMode', 'AutoencoderKL', 'CLIPTextModel', 'CLIPTextModelWithProjection', 'CacheEvictionConfig', 'ChunkStreamerBase', 'ContinuousBatchingPipeline', 'CppStdGenerator', 'DecodedResults', 'EncodedGenerationResult', 'EncodedResults', 'FluxTransformer2DModel', 'GenerationConfig', 'GenerationFinishReason', 'GenerationHandle', 'GenerationOutput', 'GenerationResult', 'GenerationStatus', 'Generator', 'Image2ImagePipeline', 'ImageGenerationConfig', 'InpaintingPipeline', 'LLMPipeline', 'MeanStdPair', 'PerfMetrics', 'PipelineMetrics', 'RawPerfMetrics', 'SD3Transformer2DModel', 'Scheduler', 'SchedulerConfig', 'StopCriteria', 'StreamerBase', 'T5EncoderModel', 'Text2ImagePipeline', 'TokenizedInputs', 'Tokenizer', 'TorchGenerator', 'UNet2DConditionModel', 'VLMDecodedResults', 'VLMPerfMetrics', 'VLMPipeline', 'VLMRawPerfMetrics', 'WhisperDecodedResultChunk', 'WhisperDecodedResults', 'WhisperGenerationConfig', 'WhisperPerfMetrics', 'WhisperPipeline', 'WhisperRawPerfMetrics', 'draft_model']
+__all__ = ['Adapter', 'AdapterConfig', 'AggregationMode', 'AutoencoderKL', 'CLIPTextModel', 'CLIPTextModelWithProjection', 'CacheEvictionConfig', 'ChunkStreamerBase', 'ContinuousBatchingPipeline', 'CppStdGenerator', 'DecodedResults', 'EncodedGenerationResult', 'EncodedResults', 'FluxTransformer2DModel', 'GenerationConfig', 'GenerationFinishReason', 'GenerationHandle', 'GenerationOutput', 'GenerationResult', 'GenerationStatus', 'Generator', 'Image2ImagePipeline', 'ImageGenerationConfig', 'InpaintingPipeline', 'LLMPipeline', 'MeanStdPair', 'PerfMetrics', 'PipelineMetrics', 'RawPerfMetrics', 'SD3Transformer2DModel', 'Scheduler', 'SchedulerConfig', 'StopCriteria', 'StreamerBase', 'T5EncoderModel', 'Text2ImagePipeline', 'TokenizedInputs', 'Tokenizer', 'TorchGenerator', 'UNet2DConditionModel', 'VLMDecodedResults', 'VLMPerfMetrics', 'VLMPipeline', 'VLMRawPerfMetrics', 'WhisperDecodedResultChunk', 'WhisperDecodedResults', 'WhisperGenerationConfig', 'WhisperPerfMetrics', 'WhisperPipeline', 'WhisperRawPerfMetrics', 'draft_model', 'get_version']
 class Adapter:
     """
     Immutable LoRA Adapter that carries the adaptation matrices and serves as unique adapter identifier.
@@ -2204,3 +2204,7 @@ def draft_model(models_path: os.PathLike, device: str = '', **kwargs) -> openvin
     """
     device on which inference will be performed
     """
+def get_version() -> str:
+    """
+    OpenVINO GenAI version
+    """
diff --git a/src/python/py_openvino_genai.cpp b/src/python/py_openvino_genai.cpp
index 429f48f30d..f8e577d5c8 100644
--- a/src/python/py_openvino_genai.cpp
+++ b/src/python/py_openvino_genai.cpp
@@ -11,6 +11,7 @@
 #include <pybind11/typing.h>
 
 #include "openvino/genai/llm_pipeline.hpp"
+#include "openvino/genai/version.hpp"
 
 #include "py_utils.hpp"
 
@@ -21,6 +22,7 @@ using ov::genai::DecodedResults;
 using ov::genai::EncodedResults;
 using ov::genai::StreamerBase;
 using ov::genai::StringInputs;
+using ov::genai::get_version;
 
 void init_lora_adapter(py::module_& m);
 void init_perf_metrics(py::module_& m);
@@ -82,7 +84,12 @@ class ConstructableStreamer: public StreamerBase {
 PYBIND11_MODULE(py_openvino_genai, m) {
     m.doc() = "Pybind11 binding for OpenVINO GenAI library";
 
+    m.def("get_version", [] () -> py::str {
+        return get_version().buildNumber;
+    }, get_version().description);
+
     init_perf_metrics(m);
+
     py::class_<DecodedResults>(m, "DecodedResults", decoded_results_docstring)
         .def(py::init<>())
         .def_property_readonly("texts", [](const DecodedResults &dr) -> py::typing::List<py::str> { return pyutils::handle_utf8((std::vector<std::string>)dr); })
diff --git a/tools/llm_bench/llm_bench_utils/config_class.py b/tools/llm_bench/llm_bench_utils/config_class.py
index 7dd27b198b..9c149c98b6 100644
--- a/tools/llm_bench/llm_bench_utils/config_class.py
+++ b/tools/llm_bench/llm_bench_utils/config_class.py
@@ -102,7 +102,9 @@
         "olmo",
         "phi3",
         "starcoder",
-        "instruct-gpt"
+        "instruct-gpt",
+        "granite",
+        "granitemoe",
     ],
     'ldm_super_resolution': ['ldm-super-resolution'],
 }
diff --git a/tools/llm_bench/llm_bench_utils/ov_utils.py b/tools/llm_bench/llm_bench_utils/ov_utils.py
index 316c9d0b89..596da8cb3a 100644
--- a/tools/llm_bench/llm_bench_utils/ov_utils.py
+++ b/tools/llm_bench/llm_bench_utils/ov_utils.py
@@ -701,7 +701,7 @@ def put(self, token_id: int) -> bool:
                 word = text[self.print_len:]
                 self.tokens_cache = []
                 self.print_len = 0
-            elif len(text) >= 3 and text[-3:] == chr(65533):
+            elif len(text) >= 3 and text[-1] == chr(65533):
                 # Don't print incomplete text.
                 pass
             elif len(text) > self.print_len:
diff --git a/tools/who_what_benchmark/whowhatbench/model_loaders.py b/tools/who_what_benchmark/whowhatbench/model_loaders.py
index 8a00c70852..c792a3c0b2 100644
--- a/tools/who_what_benchmark/whowhatbench/model_loaders.py
+++ b/tools/who_what_benchmark/whowhatbench/model_loaders.py
@@ -41,8 +41,19 @@ def load_text_genai_pipeline(model_dir, device="CPU", ov_config=None):
     return GenAIModelWrapper(openvino_genai.LLMPipeline(model_dir, device=device, **ov_config), model_dir, "text")
 
 
+def load_text_llamacpp_pipeline(model_dir):
+    try:
+        from llama_cpp import Llama
+    except ImportError:
+        logger.error(
+            "Failed to import llama_cpp package. Please install llama-cpp-python.")
+        exit(-1)
+    model = Llama(model_dir)
+    return model
+
+
 def load_text_model(
-    model_id, device="CPU", ov_config=None, use_hf=False, use_genai=False
+    model_id, device="CPU", ov_config=None, use_hf=False, use_genai=False, use_llamacpp=False,
 ):
     if use_hf:
         logger.info("Using HF Transformers API")
@@ -53,6 +64,9 @@ def load_text_model(
     elif use_genai:
         logger.info("Using OpenVINO GenAI API")
         model = load_text_genai_pipeline(model_id, device, ov_config)
+    elif use_llamacpp:
+        logger.info("Using llama.cpp API")
+        model = load_text_llamacpp_pipeline(model_id)
     else:
         logger.info("Using Optimum API")
         from optimum.intel.openvino import OVModelForCausalLM
@@ -276,7 +290,7 @@ def load_inpainting_model(
 
 
 def load_model(
-    model_type, model_id, device="CPU", ov_config=None, use_hf=False, use_genai=False
+    model_type, model_id, device="CPU", ov_config=None, use_hf=False, use_genai=False, use_llamacpp=False
 ):
     if model_id is None:
         return None
@@ -288,7 +302,7 @@ def load_model(
         ov_options = {}
 
     if model_type == "text":
-        return load_text_model(model_id, device, ov_options, use_hf, use_genai)
+        return load_text_model(model_id, device, ov_options, use_hf, use_genai, use_llamacpp)
     elif model_type == "text-to-image":
         return load_text2image_model(
             model_id, device, ov_options, use_hf, use_genai
diff --git a/tools/who_what_benchmark/whowhatbench/text_evaluator.py b/tools/who_what_benchmark/whowhatbench/text_evaluator.py
index 50ce224def..433521a186 100644
--- a/tools/who_what_benchmark/whowhatbench/text_evaluator.py
+++ b/tools/who_what_benchmark/whowhatbench/text_evaluator.py
@@ -108,6 +108,7 @@ def __init__(
         generation_config=None,
         generation_config_base=None,
         seqs_per_request=None,
+        use_chat_template=None,
     ) -> None:
         assert (
             base_model is not None or gt_data is not None
@@ -123,6 +124,7 @@ def __init__(
         self.generation_config_base = generation_config
         self.seqs_per_request = seqs_per_request
         self.generation_fn = gen_answer_fn
+        self.use_chat_template = use_chat_template
         if self.generation_config is not None:
             assert self.seqs_per_request is not None
 
@@ -202,15 +204,21 @@ def worst_examples(self, top_k: int = 5, metric="similarity"):
         return res
 
     def _generate_data(self, model, gen_answer_fn=None, generation_config=None):
-        def default_gen_answer(model, tokenizer, prompt, max_new_tokens, crop_question):
-            inputs = self.tokenizer(prompt, return_tensors="pt")
-
-            tokens = model.generate(**inputs, do_sample=False, max_new_tokens=max_new_tokens)
-
-            if crop_question:
-                tokens = tokens[:, inputs["input_ids"].shape[-1] :]
-
-            return self.tokenizer.batch_decode(tokens, skip_special_tokens=True)[0]
+        def default_gen_answer(model, tokenizer, prompt, max_new_tokens, crop_question, use_chat_template=False):
+            if use_chat_template:
+                message = [{"role": "user", "content": prompt}]
+                inputs = tokenizer.apply_chat_template(message, tokenize=True, add_generation_prompt=True, return_tensors="pt")
+                tokens = model.generate(inputs, do_sample=False, max_new_tokens=max_new_tokens)
+                if crop_question:
+                    tokens = tokens[:, inputs.shape[-1]:]
+                res = self.tokenizer.decode(tokens[0], skip_special_tokens=True)
+                return res
+            else:
+                inputs = self.tokenizer(prompt, return_tensors="pt")
+                tokens = model.generate(**inputs, do_sample=False, max_new_tokens=max_new_tokens)
+                if crop_question:
+                    tokens = tokens[:, inputs["input_ids"].shape[-1] :]
+                return self.tokenizer.batch_decode(tokens, skip_special_tokens=True)[0]
 
         gen_answer_fn = gen_answer_fn or default_gen_answer
 
@@ -250,6 +258,7 @@ def default_gen_answer(model, tokenizer, prompt, max_new_tokens, crop_question):
                         p,
                         self.max_new_tokens,
                         self._crop_question,
+                        self.use_chat_template
                     )
                 )
         else:
diff --git a/tools/who_what_benchmark/whowhatbench/wwb.py b/tools/who_what_benchmark/whowhatbench/wwb.py
index 7acf3cf5aa..7d4354f846 100644
--- a/tools/who_what_benchmark/whowhatbench/wwb.py
+++ b/tools/who_what_benchmark/whowhatbench/wwb.py
@@ -40,6 +40,11 @@ def parse_args():
         default=None,
         help="Tokenizer for divergency metric. If not provided, it will be load from base_model or target_model.",
     )
+    parser.add_argument(
+        "--chat-template",
+        action="store_true",
+        help="Whether apply the default chat template.",
+    )
     parser.add_argument(
         "--gt-data",
         default=None,
@@ -137,6 +142,11 @@ def parse_args():
         action="store_true",
         help="Use LLMPipeline from transformers library to instantiate the model.",
     )
+    parser.add_argument(
+        "--llamacpp",
+        action="store_true",
+        help="Use llama-cpp-python to instantiate the model.",
+    )
     parser.add_argument(
         "--image-size",
         type=int,
@@ -190,9 +200,13 @@ def load_prompts(args):
 def load_tokenizer(args):
     tokenizer = None
     if args.tokenizer is not None:
-        tokenizer = AutoTokenizer.from_pretrained(
-            args.tokenizer, trust_remote_code=True
-        )
+        if args.llamacpp:
+            from llama_cpp.llama_tokenizer import LlamaHFTokenizer
+            tokenizer = LlamaHFTokenizer.from_pretrained(args.tokenizer)
+        else:
+            tokenizer = AutoTokenizer.from_pretrained(
+                args.tokenizer, trust_remote_code=True
+            )
     elif args.base_model is not None:
         tokenizer = AutoTokenizer.from_pretrained(
             args.base_model, trust_remote_code=True
@@ -246,8 +260,29 @@ def diff_strings(a: str, b: str, *, use_loguru_colors: bool = False) -> str:
     return "".join(output)
 
 
-def genai_gen_text(model, tokenizer, question, max_new_tokens, skip_question):
-    return model.generate(question, do_sample=False, max_new_tokens=max_new_tokens)
+def genai_gen_text(model, tokenizer, question, max_new_tokens, skip_question, use_chat_template=False):
+    if use_chat_template:
+        model.start_chat()
+        result = model.generate(question, do_sample=False, max_new_tokens=max_new_tokens)
+        model.finish_chat()
+        return result
+    else:
+        return model.generate(question, do_sample=False, max_new_tokens=max_new_tokens)
+
+
+def llamacpp_gen_text(model, tokenizer, question, max_new_tokens, skip_question, use_chat_template=False):
+    if use_chat_template:
+        output = model.create_chat_completion(messages=[{"role": "user", "content": question}], max_tokens=max_new_tokens, temperature=0.0)
+        text = output["choices"][0]["message"]["content"]
+        if skip_question:
+            text = text[len(question):]
+        return text
+    else:
+        output = model(question, max_tokens=max_new_tokens, echo=True, temperature=0.0)
+        text = output["choices"][0]["text"]
+        if skip_question:
+            text = text[len(question):]
+        return text
 
 
 def genai_gen_image(model, prompt, num_inference_steps, generator=None):
@@ -322,7 +357,15 @@ def create_evaluator(base_model, args):
         prompts = load_prompts(args)
 
         if task == "text":
-            tokenizer = load_tokenizer(args)
+            tokenizer = load_tokenizer(args) if not args.llamacpp else None
+
+            if args.genai:
+                gen_answer_fn = genai_gen_text
+            elif args.llamacpp:
+                gen_answer_fn = llamacpp_gen_text
+            else:
+                gen_answer_fn = None
+
             return EvaluatorCLS(
                 base_model=base_model,
                 gt_data=args.gt_data,
@@ -331,7 +374,8 @@ def create_evaluator(base_model, args):
                 similarity_model_id=args.data_encoder,
                 num_samples=args.num_samples,
                 language=args.language,
-                gen_answer_fn=genai_gen_text if args.genai else None,
+                gen_answer_fn=gen_answer_fn,
+                use_chat_template=args.chat_template,
             )
         elif task == "text-to-image":
             return EvaluatorCLS(
@@ -467,10 +511,11 @@ def main():
                 args.ov_config,
                 args.hf,
                 args.genai,
+                args.llamacpp
             )
             all_metrics_per_question, all_metrics = evaluator.score(
                 target_model,
-                evaluator.get_generation_fn() if args.genai else None,
+                evaluator.get_generation_fn() if args.genai or args.llamacpp else None,
                 output_dir=args.output
             )
         logger.info("Metrics for model: %s", args.target_model)