[Frontend][last/5] Make pooling entrypoints request schema consensus. (#31127)

Signed-off-by: wang.yuqi <yuqi.wang@daocloud.io>
This commit is contained in:
wang.yuqi
2026-02-09 06:42:38 +00:00
committed by GitHub
parent 7c233dbb36
commit 22b64948f6
24 changed files with 658 additions and 612 deletions
@@ -7,12 +7,15 @@ import pytest
import torch
from tests.models.utils import softmax
from vllm import LLM, PoolingParams
from vllm import LLM, ClassificationRequestOutput, PoolingParams, PoolingRequestOutput
from vllm.distributed import cleanup_dist_env_and_memory
from vllm.tasks import PoolingTask
MODEL_NAME = "jason9693/Qwen2.5-1.5B-apeach"
prompts = ["The chef prepared a delicious meal."]
prompt = "The chef prepared a delicious meal."
prompt_token_ids = [785, 29706, 10030, 264, 17923, 15145, 13]
num_labels = 2
@pytest.fixture(scope="module")
@@ -35,11 +38,48 @@ def llm():
cleanup_dist_env_and_memory()
@pytest.mark.skip_global_cleanup
def test_str_prompts(llm: LLM):
outputs = llm.classify(prompt, use_tqdm=False)
assert len(outputs) == 1
assert isinstance(outputs[0], ClassificationRequestOutput)
assert outputs[0].prompt_token_ids == prompt_token_ids
assert len(outputs[0].outputs.probs) == num_labels
@pytest.mark.skip_global_cleanup
def test_token_ids_prompts(llm: LLM):
outputs = llm.classify([prompt_token_ids], use_tqdm=False)
assert len(outputs) == 1
assert isinstance(outputs[0], ClassificationRequestOutput)
assert outputs[0].prompt_token_ids == prompt_token_ids
assert len(outputs[0].outputs.probs) == num_labels
@pytest.mark.skip_global_cleanup
def test_list_prompts(llm: LLM):
outputs = llm.classify([prompt, prompt_token_ids], use_tqdm=False)
assert len(outputs) == 2
for i in range(len(outputs)):
assert isinstance(outputs[i], ClassificationRequestOutput)
assert outputs[i].prompt_token_ids == prompt_token_ids
assert len(outputs[i].outputs.probs) == num_labels
@pytest.mark.skip_global_cleanup
def test_token_classify(llm: LLM):
outputs = llm.encode(prompt, pooling_task="token_classify", use_tqdm=False)
assert len(outputs) == 1
assert isinstance(outputs[0], PoolingRequestOutput)
assert outputs[0].prompt_token_ids == prompt_token_ids
assert outputs[0].outputs.data.shape == (len(prompt_token_ids), num_labels)
@pytest.mark.skip_global_cleanup
def test_pooling_params(llm: LLM):
def get_outputs(use_activation):
outputs = llm.classify(
prompts,
prompt,
pooling_params=PoolingParams(use_activation=use_activation),
use_tqdm=False,
)
@@ -61,11 +101,14 @@ def test_pooling_params(llm: LLM):
@pytest.mark.skip_global_cleanup
def test_token_classify(llm: LLM):
llm.encode(prompts, pooling_task="token_classify", use_tqdm=False)
def test_score_api(llm: LLM):
err_msg = "Score API is only enabled for num_labels == 1."
with pytest.raises(ValueError, match=err_msg):
llm.score("ping", "pong", use_tqdm=False)
@pytest.mark.parametrize("task", ["embed", "token_embed", "plugin"])
def test_unsupported_tasks(llm: LLM, task: PoolingTask):
err_msg = f"Unsupported task: '{task}' Supported tasks.+"
with pytest.raises(ValueError, match=err_msg):
llm.encode(prompt, pooling_task=task, use_tqdm=False)
@@ -10,12 +10,12 @@ from transformers import AutoProcessor
from tests.utils import VLLM_PATH, RemoteOpenAIServer
from vllm.entrypoints.pooling.embed.protocol import EmbeddingResponse
from vllm.multimodal.media import MediaWithBytes
from vllm.multimodal.utils import fetch_image
from vllm.multimodal.utils import encode_image_url, fetch_image
MODEL_NAME = "TIGER-Lab/VLM2Vec-Full"
MAXIMUM_IMAGES = 2
vlm2vec_jinja_path = VLLM_PATH / "examples/template_vlm2vec_phi3v.jinja"
vlm2vec_jinja_path = VLLM_PATH / "examples/pooling/embed/template/vlm2vec_phi3v.jinja"
assert vlm2vec_jinja_path.exists()
# Test different image extensions (JPG/PNG) and formats (gray/RGB/RGBA)
@@ -26,6 +26,10 @@ TEST_IMAGE_ASSETS = [
"RGBA_comp.png", # "https://vllm-public-assets.s3.us-west-2.amazonaws.com/vision_model_images/RGBA_comp.png",
]
input_text = "The best thing about vLLM is that it supports many different models"
image_url = "https://vllm-public-assets.s3.us-west-2.amazonaws.com/multimodal_asset/cat_snow.jpg"
image_base64 = {"url": encode_image_url(fetch_image(image_url))}
@pytest.fixture(scope="module")
def server():
@@ -48,6 +52,81 @@ def server():
yield remote_server
@pytest.mark.parametrize("model_name", [MODEL_NAME])
def test_chat_text_request(server: RemoteOpenAIServer, model_name: str):
messages = [
{
"role": "user",
"content": input_text,
},
]
# note: vlm2vec_phi3v.jinja
# Embedding models should only embed one message at a time.
response = requests.post(
server.url_for("v1/embeddings"),
json={"model": model_name, "messages": messages},
)
response.raise_for_status()
output = EmbeddingResponse.model_validate(response.json())
assert len(output.data) == 1
assert output.model == MODEL_NAME
assert len(output.data[0].embedding) == 3072
assert output.usage.prompt_tokens == 14
@pytest.mark.parametrize("model_name", [MODEL_NAME])
def test_chat_image_url_request(server: RemoteOpenAIServer, model_name: str):
messages = [
{
"role": "user",
"content": [
{"type": "text", "text": "Represent the user's input."},
{"type": "image_url", "image_url": {"url": image_url}},
],
}
]
response = requests.post(
server.url_for("v1/embeddings"),
json={"model": model_name, "messages": messages},
)
response.raise_for_status()
output = EmbeddingResponse.model_validate(response.json())
assert len(output.data) == 1
assert output.model == MODEL_NAME
assert len(output.data[0].embedding) == 3072
assert output.usage.prompt_tokens == 767
@pytest.mark.parametrize("model_name", [MODEL_NAME])
def test_chat_image_base64_request(server: RemoteOpenAIServer, model_name: str):
messages = [
{
"role": "user",
"content": [
{"type": "text", "text": "Represent the user's input."},
{"type": "image_url", "image_url": image_base64},
],
}
]
response = requests.post(
server.url_for("v1/embeddings"),
json={"model": model_name, "messages": messages},
)
response.raise_for_status()
output = EmbeddingResponse.model_validate(response.json())
assert len(output.data) == 1
assert output.model == MODEL_NAME
assert len(output.data[0].embedding) == 3072
assert output.usage.prompt_tokens == 767
def get_hf_prompt_tokens(model_name, content, image_url):
processor = AutoProcessor.from_pretrained(
model_name, trust_remote_code=True, num_crops=4