biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

Support embedding models in V1 (#16188)

Browse Source 
Signed-off-by: Max de Bayser <mbayser@br.ibm.com>
Signed-off-by: Max de Bayser <maxdebayser@gmail.com>
Signed-off-by: 22quinn <33176974+22quinn@users.noreply.github.com>
Co-authored-by: 22quinn <33176974+22quinn@users.noreply.github.com>
This commit is contained in:
Maximilien de Bayser
2025-06-19 01:36:33 -03:00
committed by GitHub
parent 4959915089
commit 799397ee4f
56 changed files with 889 additions and 281 deletions
Download Patch File Download Diff File Expand all files Collapse all files

34
tests/models/language/pooling/test_embedding.py
View File

@@ -8,6 +8,14 @@ from vllm.platforms import current_platform
from ...utils import check_embeddings_close
@pytest.fixture(autouse=True)
def v1(run_with_both_engines):
# Simple autouse wrapper to run both engines for each test
# This can be promoted up to conftest.py to run for every
# test in a package
pass
@pytest.mark.parametrize(
"model",
[
@@ -20,15 +28,27 @@ from ...utils import check_embeddings_close
marks=[pytest.mark.core_model]),
pytest.param("intfloat/e5-mistral-7b-instruct",
marks=[pytest.mark.core_model, pytest.mark.cpu_model]),
pytest.param("ssmits/Qwen2-7B-Instruct-embed-base"),
# the qwen models interfere with each other (see PR
# https://github.com/vllm-project/vllm/pull/18720).
# To avoid this problem, for now we skip v0 since it will be
# deprecated anyway.
pytest.param("ssmits/Qwen2-7B-Instruct-embed-base",
marks=[pytest.mark.skip_v0]),
# [Encoder-only]
pytest.param("BAAI/bge-base-en-v1.5",
marks=[pytest.mark.core_model, pytest.mark.cpu_model]),
pytest.param("sentence-transformers/all-MiniLM-L12-v2"),
pytest.param("intfloat/multilingual-e5-small"),
pytest.param("Alibaba-NLP/gte-Qwen2-1.5B-instruct"),
marks=[
pytest.mark.core_model, pytest.mark.cpu_model,
pytest.mark.skip_v1
]),
pytest.param("sentence-transformers/all-MiniLM-L12-v2",
marks=[pytest.mark.skip_v1]),
pytest.param("intfloat/multilingual-e5-small",
marks=[pytest.mark.skip_v1]),
pytest.param("Alibaba-NLP/gte-Qwen2-1.5B-instruct",
marks=[pytest.mark.skip_v1]),
# [Cross-Encoder]
pytest.param("sentence-transformers/stsb-roberta-base-v2"),
pytest.param("sentence-transformers/stsb-roberta-base-v2",
marks=[pytest.mark.skip_v1]),
],
)
def test_models(
@@ -62,7 +82,7 @@ def test_models(
with vllm_runner(model,
task="embed",
max_model_len=None,
max_model_len=512,
**vllm_extra_kwargs) as vllm_model:
vllm_outputs = vllm_model.encode(example_prompts)