[Misc] Introduce encode_*_url utility function (#31208)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>

tests/entrypoints/openai/test_audio.py

@@ -8,7 +8,7 @@ import pytest
 import pytest_asyncio
 
 from vllm.assets.audio import AudioAsset
-from vllm.multimodal.utils import encode_audio_base64, fetch_audio
+from vllm.multimodal.utils import encode_audio_base64, encode_audio_url, fetch_audio
 
 from ...utils import RemoteOpenAIServer
 
@@ -53,6 +53,14 @@ def base64_encoded_audio() -> dict[str, str]:
     }
 
 
+@pytest.fixture(scope="session")
+def url_encoded_audio() -> dict[str, str]:
+    return {
+        audio_url: encode_audio_url(*fetch_audio(audio_url))
+        for audio_url in TEST_AUDIO_URLS
+    }
+
+
 def dummy_messages_from_audio_url(
     audio_urls: str | list[str],
     content_text: str = "What's happening in this audio?",
@@ -149,11 +157,9 @@ async def test_single_chat_session_audio_base64encoded(
     client: openai.AsyncOpenAI,
     model_name: str,
     audio_url: str,
-    base64_encoded_audio: dict[str, str],
+    url_encoded_audio: dict[str, str],
 ):
-    messages = dummy_messages_from_audio_url(
-        f"data:audio/wav;base64,{base64_encoded_audio[audio_url]}"
-    )
+    messages = dummy_messages_from_audio_url(url_encoded_audio[audio_url])
 
     # test single completion
     chat_completion = await client.chat.completions.create(

tests/entrypoints/openai/test_video.py

@@ -7,7 +7,7 @@ import openai
 import pytest
 import pytest_asyncio
 
-from vllm.multimodal.utils import encode_video_base64, fetch_video
+from vllm.multimodal.utils import encode_video_url, fetch_video
 
 from ...utils import RemoteOpenAIServer
 
@@ -48,9 +48,9 @@ async def client(server):
 
 
 @pytest.fixture(scope="session")
-def base64_encoded_video() -> dict[str, str]:
+def url_encoded_video() -> dict[str, str]:
     return {
-        video_url: encode_video_base64(fetch_video(video_url)[0])
+        video_url: encode_video_url(fetch_video(video_url)[0])
         for video_url in TEST_VIDEO_URLS
     }
 
@@ -175,11 +175,9 @@ async def test_single_chat_session_video_base64encoded(
     client: openai.AsyncOpenAI,
     model_name: str,
     video_url: str,
-    base64_encoded_video: dict[str, str],
+    url_encoded_video: dict[str, str],
 ):
-    messages = dummy_messages_from_video_url(
-        f"data:video/jpeg;base64,{base64_encoded_video[video_url]}"
-    )
+    messages = dummy_messages_from_video_url(url_encoded_video[video_url])
 
     # test single completion
     chat_completion = await client.chat.completions.create(
@@ -223,11 +221,9 @@ async def test_single_chat_session_video_base64encoded_beamsearch(
     client: openai.AsyncOpenAI,
     model_name: str,
     video_url: str,
-    base64_encoded_video: dict[str, str],
+    url_encoded_video: dict[str, str],
 ):
-    messages = dummy_messages_from_video_url(
-        f"data:video/jpeg;base64,{base64_encoded_video[video_url]}"
-    )
+    messages = dummy_messages_from_video_url(url_encoded_video[video_url])
 
     chat_completion = await client.chat.completions.create(
         model=model_name,

tests/entrypoints/openai/test_vision.py

@@ -9,7 +9,7 @@ import pytest_asyncio
 from transformers import AutoProcessor
 
 from vllm.multimodal.base import MediaWithBytes
-from vllm.multimodal.utils import encode_image_base64, fetch_image
+from vllm.multimodal.utils import encode_image_url, fetch_image
 
 from ...utils import RemoteOpenAIServer
 
@@ -35,7 +35,7 @@ EXPECTED_MM_BEAM_SEARCH_RES = [
     ],
     [
         "The image shows a Venn diagram with three over",
-        "The image shows a colorful Venn diagram with",
+        "The image displays a Venn diagram with three over",
     ],
     [
         "This image displays a gradient of colors ranging from",
@@ -70,11 +70,9 @@ async def client(server):
 
 
 @pytest.fixture(scope="session")
-def base64_encoded_image(local_asset_server) -> dict[str, str]:
+def url_encoded_image(local_asset_server) -> dict[str, str]:
     return {
-        image_asset: encode_image_base64(
-            local_asset_server.get_image_asset(image_asset)
-        )
+        image_asset: encode_image_url(local_asset_server.get_image_asset(image_asset))
         for image_asset in TEST_IMAGE_ASSETS
     }
 
@@ -234,11 +232,11 @@ async def test_single_chat_session_image_base64encoded(
     model_name: str,
     raw_image_url: str,
     image_url: str,
-    base64_encoded_image: dict[str, str],
+    url_encoded_image: dict[str, str],
 ):
     content_text = "What's in this image?"
     messages = dummy_messages_from_image_url(
-        f"data:image/jpeg;base64,{base64_encoded_image[raw_image_url]}",
+        url_encoded_image[raw_image_url],
         content_text,
     )
 
@@ -288,15 +286,13 @@ async def test_single_chat_session_image_base64encoded_beamsearch(
     client: openai.AsyncOpenAI,
     model_name: str,
     image_idx: int,
-    base64_encoded_image: dict[str, str],
+    url_encoded_image: dict[str, str],
 ):
     # NOTE: This test also validates that we pass MM data through beam search
     raw_image_url = TEST_IMAGE_ASSETS[image_idx]
     expected_res = EXPECTED_MM_BEAM_SEARCH_RES[image_idx]
 
-    messages = dummy_messages_from_image_url(
-        f"data:image/jpeg;base64,{base64_encoded_image[raw_image_url]}"
-    )
+    messages = dummy_messages_from_image_url(url_encoded_image[raw_image_url])
 
     chat_completion = await client.chat.completions.create(
         model=model_name,

tests/entrypoints/pooling/embed/test_online_vision.py

@@ -10,7 +10,7 @@ from transformers import AutoProcessor
 from tests.utils import VLLM_PATH, RemoteOpenAIServer
 from vllm.entrypoints.pooling.embed.protocol import EmbeddingResponse
 from vllm.multimodal.base import MediaWithBytes
-from vllm.multimodal.utils import encode_image_base64, fetch_image
+from vllm.multimodal.utils import fetch_image
 
 MODEL_NAME = "TIGER-Lab/VLM2Vec-Full"
 MAXIMUM_IMAGES = 2
@@ -48,14 +48,6 @@ def server():
         yield remote_server
 
 
-@pytest.fixture(scope="session")
-def base64_encoded_image(local_asset_server) -> dict[str, str]:
-    return {
-        image_url: encode_image_base64(local_asset_server.get_image_asset(image_url))
-        for image_url in TEST_IMAGE_ASSETS
-    }
-
-
 def get_hf_prompt_tokens(model_name, content, image_url):
     processor = AutoProcessor.from_pretrained(
         model_name, trust_remote_code=True, num_crops=4

tests/entrypoints/test_chat_utils.py

@@ -25,9 +25,9 @@ from vllm.entrypoints.chat_utils import (
 )
 from vllm.multimodal import MultiModalDataDict, MultiModalUUIDDict
 from vllm.multimodal.utils import (
-    encode_audio_base64,
-    encode_image_base64,
-    encode_video_base64,
+    encode_audio_url,
+    encode_image_url,
+    encode_video_url,
 )
 from vllm.tokenizers import get_tokenizer
 from vllm.tokenizers.mistral import MistralTokenizer
@@ -141,22 +141,19 @@ def mistral_model_config():
 @pytest.fixture(scope="module")
 def image_url():
     image = ImageAsset("cherry_blossom")
-    base64 = encode_image_base64(image.pil_image)
-    return f"data:image/jpeg;base64,{base64}"
+    return encode_image_url(image.pil_image)
 
 
 @pytest.fixture(scope="module")
 def video_url():
     video = VideoAsset("baby_reading", 1)
-    base64 = encode_video_base64(video.np_ndarrays)
-    return f"data:video/jpeg;base64,{base64}"
+    return encode_video_url(video.np_ndarrays)
 
 
 @pytest.fixture(scope="module")
 def audio_url():
     audio = AudioAsset("mary_had_lamb")
-    base64 = encode_audio_base64(*audio.audio_and_sample_rate)
-    return f"data:audio/ogg;base64,{base64}"
+    return encode_audio_url(*audio.audio_and_sample_rate)
 
 
 def _assert_mm_data_is_image_input(