[GHA] Replaced visual_language_chat_sample-ubuntu-llava (#1802)

Author: mryzhov

.github/workflows/causal_lm_cpp.yml

@@ -363,18 +363,6 @@ jobs:
             f.write(content.encode("utf-8"))
       - run: diff cpp2.txt py2.txt
 
-  visual_language_chat_sample-ubuntu-llava_1_5:
-    uses: ./.github/workflows/job_vlm_sample_llava.yml
-    with:
-      model_id: llava-hf/llava-1.5-7b-hf
-      model_dir: llava_1_5_7b_ov
-
-  visual_language_chat_sample-ubuntu-llava_next:
-    uses: ./.github/workflows/job_vlm_sample_llava.yml
-    with:
-      model_id: llava-hf/llava-v1.6-mistral-7b-hf
-      model_dir: llava_v1_6_mistral_7b_ov
-
   visual_language_chat_sample-ubuntu-internvl2:
     runs-on: ubuntu-22.04-16-cores
     steps:
@@ -575,7 +563,7 @@ jobs:
   Overall_Status:
     name: ci/gha_overall_status_causal_lm
     needs: [cpp-greedy_causal_lm-windows, cpp-Phi-1_5, cpp-greedy_causal_lm-redpajama-3b-chat, cpp-chat_sample-ubuntu, cpp-continuous-batching-ubuntu,
-            visual_language_chat_sample-ubuntu-minicpm_v2_6, visual_language_chat_sample-ubuntu-llava_1_5, visual_language_chat_sample-ubuntu-llava_next, visual_language_chat_sample-ubuntu-internvl2,
+            visual_language_chat_sample-ubuntu-minicpm_v2_6, visual_language_chat_sample-ubuntu-internvl2,
             cpp-continuous-batching-windows, cpp-continuous-batching-macos]
     if: ${{ always() }}
     runs-on: ubuntu-latest

.github/workflows/job_vlm_sample_llava.yml

@@ -336,9 +336,13 @@ jobs:
             marker: 'image_generation'
             cmd: 'tests/python_tests/samples'
             runner: 'aks-linux-8-cores-64gb'
+          - name: 'VLM'
+            marker: 'vlm'
+            cmd: 'tests/python_tests/samples'
+            runner: 'aks-linux-8-cores-64gb'
 
     needs: [ openvino_download, genai_build_cmake, genai_build_wheel, genai_build_samples ] 
-    timeout-minutes: 45
+    timeout-minutes: 60
     defaults:
       run:
         shell: bash

.github/workflows/linux.yml

@@ -9,12 +9,14 @@ markers =
     ; llm         - Tests related to large language models.
     ; whisper     - Tests related to the Whisper model.
     ; image_generation - Tests related to image generation.
+    ; vlm         - Tests related to the VLM model.
     precommit
     nightly
     real_models
     samples
     llm
     whisper
     image_generation
+    vlm
 
 addopts = -m precommit

tests/python_tests/pytest.ini

@@ -57,6 +57,22 @@
         "name": "dreamlike-art/dreamlike-anime-1.0",
         "convert_args": ['--trust-remote-code', '--weight-format', 'fp16', "--task", "stable-diffusion"]
     },
+    "LCM_Dreamshaper_v7-int8-ov": {
+        "name": "OpenVINO/LCM_Dreamshaper_v7-int8-ov",
+        "convert_args": []
+    },   
+    "llava-1.5-7b-hf": {
+        "name": "llava-hf/llava-1.5-7b-hf",
+        "convert_args": ['--trust-remote-code']
+    },    
+    "llava-v1.6-mistral-7b-hf": {
+        "name": "llava-hf/llava-v1.6-mistral-7b-hf",
+        "convert_args": ['--trust-remote-code']
+    },
+    "dreamlike-anime-1.0": {
+        "name": "dreamlike-art/dreamlike-anime-1.0",
+        "convert_args": ['--trust-remote-code', '--weight-format', 'fp16', "--task", "stable-diffusion"]
+    },
     "LCM_Dreamshaper_v7-int8-ov": {
         "name": "OpenVINO/LCM_Dreamshaper_v7-int8-ov",
         "convert_args": []
@@ -66,6 +82,7 @@
 TEST_FILES = {
     "how_are_you_doing_today.wav": "https://storage.openvinotoolkit.org/models_contrib/speech/2021.2/librispeech_s5/how_are_you_doing_today.wav",
     "adapter_model.safetensors": "https://huggingface.co/smangrul/tinyllama_lora_sql/resolve/main/adapter_model.safetensors",
+    "monalisa.jpg": "https://llava-vl.github.io/static/images/monalisa.jpg",
     "soulcard.safetensors": "https://civitai.com/api/download/models/72591",
     "image.png": "https://raw.githubusercontent.com/CompVis/latent-diffusion/main/data/inpainting_examples/overture-creations-5sI6fQgYIuo.png",
     "mask_image.png": "https://raw.githubusercontent.com/CompVis/latent-diffusion/main/data/inpainting_examples/overture-creations-5sI6fQgYIuo_mask.png"

tests/python_tests/samples/conftest.py

@@ -3,8 +3,10 @@
 from conftest import logger
 import subprocess # nosec B404
 
-def run_sample(command):
+def run_sample(command, input_data=None):
     logger.info(f"Running sample command: {' '.join(command)}")
-    result = subprocess.run(command, capture_output=True, text=True, check=True, encoding='utf-8')
+    if input_data:
+        logger.info(f"Input data: {input_data}")
+    result = subprocess.run(command, capture_output=True, text=True, check=True, encoding='utf-8', input=input_data)
     logger.info(f"Sample output: {result.stdout}")
     return result

tests/python_tests/samples/test_utils.py

@@ -0,0 +1,35 @@
+# Copyright (C) 2025 Intel Corporation
+# SPDX-License-Identifier: Apache-2.0
+
+import os
+import pytest
+import sys
+
+from conftest import SAMPLES_PY_DIR, SAMPLES_CPP_DIR
+from test_utils import run_sample
+
+class TestVisualLanguageChat:
+    @pytest.mark.vlm
+    @pytest.mark.samples
+    @pytest.mark.parametrize(
+        "convert_model, sample_args",
+        [
+            pytest.param("llava-1.5-7b-hf", 'Who drew this painting?\nWhen did the painter live?'),
+            pytest.param("llava-v1.6-mistral-7b-hf", 'Who drew this painting?\nWhen did the painter live?'),
+        ],
+        indirect=["convert_model"],
+    )
+    @pytest.mark.parametrize("download_test_content", ["monalisa.jpg"], indirect=True)
+    def test_sample_visual_language_chat(self, convert_model, download_test_content, sample_args):
+        # Test Python sample
+        py_script = os.path.join(SAMPLES_PY_DIR, "visual_language_chat/visual_language_chat.py")
+        py_command = [sys.executable, py_script, convert_model, download_test_content]
+        py_result = run_sample(py_command, sample_args)
+
+        # Test CPP sample
+        cpp_sample = os.path.join(SAMPLES_CPP_DIR, 'visual_language_chat')
+        cpp_command =[cpp_sample, convert_model, download_test_content]
+        cpp_result = run_sample(cpp_command, sample_args)
+
+        # Compare results
+        assert py_result.stdout == cpp_result.stdout, f"Results should match"