[CI] Mirror external test assets in vLLM S3 (#52064)

Signed-off-by: khluu <khluu000@gmail.com>
Co-authored-by: OpenAI Codex <noreply@openai.com>

tests/entrypoints/multimodal/openai/chat_completion/test_video.py

@@ -8,6 +8,7 @@ import pytest
 import pytest_asyncio
 
 from tests.utils import RemoteOpenAIServer
+from vllm.assets.base import VLLM_S3_BUCKET_URL
 from vllm.multimodal.utils import encode_video_url, fetch_video
 from vllm.platforms import current_platform
 
@@ -15,9 +16,9 @@ MODEL_NAME = "llava-hf/llava-onevision-qwen2-0.5b-ov-hf"
 MAXIMUM_VIDEOS = 3
 
 TEST_VIDEO_URLS = [
-    "https://www.bogotobogo.com/python/OpenCV_Python/images/mean_shift_tracking/slow_traffic_small.mp4",
-    "https://github.com/opencv/opencv/raw/refs/tags/4.12.0/samples/data/vtest.avi",
-    "https://github.com/opencv/opencv/raw/refs/tags/4.12.0/samples/data/Megamind.avi",
+    f"{VLLM_S3_BUCKET_URL}/multimodal_asset/slow_traffic_small.mp4",
+    f"{VLLM_S3_BUCKET_URL}/multimodal_asset/vtest.avi",
+    f"{VLLM_S3_BUCKET_URL}/multimodal_asset/Megamind.avi",
 ]
 
 

tests/entrypoints/pooling/classify/test_online_vision.py

@@ -6,6 +6,7 @@ import pytest
 import requests
 
 from tests.utils import RemoteOpenAIServer
+from vllm.assets.base import VLLM_S3_BUCKET_URL
 from vllm.entrypoints.pooling.classify.protocol import ClassificationResponse
 from vllm.multimodal.utils import encode_image_url, fetch_image
 
@@ -14,9 +15,9 @@ MAXIMUM_VIDEOS = 1
 
 HF_OVERRIDES = {"architectures": ["Qwen2_5_VLForSequenceClassification"]}
 input_text = "This product was excellent and exceeded my expectations"
-image_url = "https://vllm-public-assets.s3.us-west-2.amazonaws.com/multimodal_asset/cat_snow.jpg"
+image_url = f"{VLLM_S3_BUCKET_URL}/multimodal_asset/cat_snow.jpg"
 image_base64 = {"url": encode_image_url(fetch_image(image_url))}
-video_url = "https://www.bogotobogo.com/python/OpenCV_Python/images/mean_shift_tracking/slow_traffic_small.mp4"
+video_url = f"{VLLM_S3_BUCKET_URL}/multimodal_asset/slow_traffic_small.mp4"
 
 
 @pytest.fixture(scope="module")

tests/evals/gsm8k/gsm8k_eval.py

@@ -20,6 +20,8 @@ import regex as re
 import requests
 from tqdm.asyncio import tqdm
 
+from vllm.assets.base import VLLM_S3_BUCKET_URL
+
 INVALID = -9999999
 
 
@@ -44,8 +46,8 @@ def download_and_cache_file(url: str, filename: str | None = None) -> str:
 
 def load_gsm8k_data() -> tuple[list[dict], list[dict]]:
     """Load GSM8K train and test data"""
-    train_url = "https://raw.githubusercontent.com/openai/grade-school-math/master/grade_school_math/data/train.jsonl"
-    test_url = "https://raw.githubusercontent.com/openai/grade-school-math/master/grade_school_math/data/test.jsonl"
+    train_url = f"{VLLM_S3_BUCKET_URL}/ci-datasets/gsm8k/train.jsonl"
+    test_url = f"{VLLM_S3_BUCKET_URL}/ci-datasets/gsm8k/test.jsonl"
 
     train_file = download_and_cache_file(train_url)
     test_file = download_and_cache_file(test_url)

tests/models/multimodal/pooling/test_jinavl_reranker.py

@@ -6,6 +6,7 @@ import transformers
 from packaging import version
 from transformers import AutoModel
 
+from vllm.assets.base import VLLM_S3_BUCKET_URL
 from vllm.entrypoints.chat_utils import (
     ChatCompletionContentPartImageEmbedsParam,
     ChatCompletionContentPartImageParam,
@@ -34,6 +35,11 @@ CHECKPOINT_TO_HF_MAPPER = {
     "model.": "model.language_model.",
 }
 
+HANDELSBLATT_IMAGE_URL = (
+    f"{VLLM_S3_BUCKET_URL}/multimodal_asset/jinavl-handelsblatt-preview.png"
+)
+PAPER_IMAGE_URL = f"{VLLM_S3_BUCKET_URL}/multimodal_asset/jinavl-paper-11.png"
+
 # Shared long text for test data
 LONG_TEXT_DOC = """We present ReaderLM-v2, a compact 1.5 billion parameter language model designed for efficient
 web content extraction. Our model processes documents up to 512K tokens, transforming messy HTML
@@ -50,12 +56,8 @@ lower computational requirements."""  # noqa: E501
 TEXT_IMAGE_TEST_DATA = {
     "query": [{"text": "slm markdown"}],
     "documents": [
-        {
-            "image": "https://raw.githubusercontent.com/jina-ai/multimodal-reranker-test/main/handelsblatt-preview.png"
-        },
-        {
-            "image": "https://raw.githubusercontent.com/jina-ai/multimodal-reranker-test/main/paper-11.png"
-        },
+        {"image": HANDELSBLATT_IMAGE_URL},
+        {"image": PAPER_IMAGE_URL},
     ],
 }
 
@@ -68,11 +70,7 @@ TEXT_TEXT_TEST_DATA = {
 }
 
 IMAGE_TEXT_TEST_DATA = {
-    "query": [
-        {
-            "image": "https://raw.githubusercontent.com/jina-ai/multimodal-reranker-test/main/paper-11.png"
-        }
-    ],
+    "query": [{"image": PAPER_IMAGE_URL}],
     "documents": [
         {"text": LONG_TEXT_DOC},
         {"text": "数据提取么?为什么不用正则啊,你用正则不就全解决了么?"},
@@ -80,18 +78,10 @@ IMAGE_TEXT_TEST_DATA = {
 }
 
 IMAGE_IMAGE_TEST_DATA = {
-    "query": [
-        {
-            "image": "https://raw.githubusercontent.com/jina-ai/multimodal-reranker-test/main/paper-11.png"
-        }
-    ],
+    "query": [{"image": PAPER_IMAGE_URL}],
     "documents": [
-        {
-            "image": "https://raw.githubusercontent.com/jina-ai/multimodal-reranker-test/main/handelsblatt-preview.png"
-        },
-        {
-            "image": "https://raw.githubusercontent.com/jina-ai/multimodal-reranker-test/main/paper-11.png"
-        },
+        {"image": HANDELSBLATT_IMAGE_URL},
+        {"image": PAPER_IMAGE_URL},
     ],
 }
 
@@ -99,13 +89,9 @@ TEXT_MIXED_DOCS_TEST_DATA = {
     "query": [{"text": "slm markdown"}],
     "documents": [
         {"text": LONG_TEXT_DOC},
-        {
-            "image": "https://raw.githubusercontent.com/jina-ai/multimodal-reranker-test/main/paper-11.png"
-        },
+        {"image": PAPER_IMAGE_URL},
         {"text": "数据提取么？为什么不用正则啊,你用正则不就全解决了么?"},
-        {
-            "image": "https://raw.githubusercontent.com/jina-ai/multimodal-reranker-test/main/handelsblatt-preview.png"
-        },
+        {"image": HANDELSBLATT_IMAGE_URL},
     ],
 }
 

tests/multimodal/media/test_connector.py

@@ -17,6 +17,7 @@ import requests
 import torch
 from PIL import Image, ImageChops
 
+from vllm.assets.base import VLLM_S3_BUCKET_URL
 from vllm.multimodal.image import convert_image_mode
 from vllm.multimodal.inputs import PlaceholderRange
 from vllm.multimodal.media import MediaConnector
@@ -30,8 +31,8 @@ TEST_IMAGE_ASSETS = [
 ]
 
 TEST_VIDEO_URLS = [
-    "https://www.bogotobogo.com/python/OpenCV_Python/images/mean_shift_tracking/slow_traffic_small.mp4",
-    "https://github.com/opencv/opencv/raw/refs/tags/4.12.0/samples/data/vtest.avi",
+    f"{VLLM_S3_BUCKET_URL}/multimodal_asset/slow_traffic_small.mp4",
+    f"{VLLM_S3_BUCKET_URL}/multimodal_asset/vtest.avi",
 ]
 
 
@@ -76,8 +77,7 @@ async def test_fetch_image_base64(
     connector = MediaConnector(
         # Domain restriction should not apply to data URLs.
         allowed_media_domains=[
-            "www.bogotobogo.com",
-            "github.com",
+            VLLM_S3_BUCKET_URL.removeprefix("https://"),
         ]
     )
     url_image = url_images[raw_image_url]
@@ -390,8 +390,7 @@ async def test_allowed_media_domains(video_url: str, num_frames: int):
             }
         },
         allowed_media_domains=[
-            "www.bogotobogo.com",
-            "github.com",
+            VLLM_S3_BUCKET_URL.removeprefix("https://"),
         ],
     )