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

Convert formatting to use ruff instead of yapf + isort (#26247)

Browse Source 
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
This commit is contained in:
Harry Mellor
2025-10-05 15:06:22 +01:00
committed by GitHub
parent 17edd8a807
commit d6953beb91
1508 changed files with 115244 additions and 94146 deletions
Download Patch File Download Diff File Expand all files Collapse all files

86
tests/models/language/pooling_mteb_test/test_jina.py
View File

@@ -5,60 +5,68 @@ from functools import partial
import pytest
from tests.models.language.pooling.embed_utils import (
check_embeddings_close, correctness_test_embed_models, matryoshka_fy)
from tests.models.utils import (CLSPoolingEmbedModelInfo,
CLSPoolingRerankModelInfo, EmbedModelInfo,
RerankModelInfo)
check_embeddings_close,
correctness_test_embed_models,
matryoshka_fy,
)
from tests.models.utils import (
CLSPoolingEmbedModelInfo,
CLSPoolingRerankModelInfo,
EmbedModelInfo,
RerankModelInfo,
)
from vllm import PoolingParams
from .mteb_utils import mteb_test_embed_models, mteb_test_rerank_models
EMBEDDING_MODELS = [
CLSPoolingEmbedModelInfo("jinaai/jina-embeddings-v3",
mteb_score=0.824413164,
architecture="XLMRobertaModel",
is_matryoshka=True)
CLSPoolingEmbedModelInfo(
"jinaai/jina-embeddings-v3",
mteb_score=0.824413164,
architecture="XLMRobertaModel",
is_matryoshka=True,
)
]
RERANK_MODELS = [
CLSPoolingRerankModelInfo(
"jinaai/jina-reranker-v2-base-multilingual",
mteb_score=0.33643,
architecture="XLMRobertaForSequenceClassification")
architecture="XLMRobertaForSequenceClassification",
)
]
@pytest.mark.parametrize("model_info", EMBEDDING_MODELS)
def test_embed_models_mteb(hf_runner, vllm_runner,
model_info: EmbedModelInfo) -> None:
def test_embed_models_mteb(hf_runner, vllm_runner, model_info: EmbedModelInfo) -> None:
def hf_model_callback(model):
model.encode = partial(model.encode, task="text-matching")
mteb_test_embed_models(hf_runner,
vllm_runner,
model_info,
hf_model_callback=hf_model_callback)
mteb_test_embed_models(
hf_runner, vllm_runner, model_info, hf_model_callback=hf_model_callback
)
@pytest.mark.parametrize("model_info", EMBEDDING_MODELS)
def test_embed_models_correctness(hf_runner, vllm_runner,
model_info: EmbedModelInfo,
example_prompts) -> None:
def test_embed_models_correctness(
hf_runner, vllm_runner, model_info: EmbedModelInfo, example_prompts
) -> None:
def hf_model_callback(model):
model.encode = partial(model.encode, task="text-matching")
correctness_test_embed_models(hf_runner,
vllm_runner,
model_info,
example_prompts,
hf_model_callback=hf_model_callback)
correctness_test_embed_models(
hf_runner,
vllm_runner,
model_info,
example_prompts,
hf_model_callback=hf_model_callback,
)
@pytest.mark.parametrize("model_info", RERANK_MODELS)
def test_rerank_models_mteb(hf_runner, vllm_runner,
model_info: RerankModelInfo) -> None:
def test_rerank_models_mteb(
hf_runner, vllm_runner, model_info: RerankModelInfo
) -> None:
mteb_test_rerank_models(hf_runner, vllm_runner, model_info)
@@ -81,32 +89,32 @@ def test_matryoshka(
example_prompts = [str(s).strip() for s in example_prompts]
with hf_runner(
model_info.name,
dtype=dtype,
is_sentence_transformer=True,
model_info.name,
dtype=dtype,
is_sentence_transformer=True,
) as hf_model:
hf_outputs = hf_model.encode(example_prompts, task="text-matching")
hf_outputs = matryoshka_fy(hf_outputs, dimensions)
with vllm_runner(model_info.name,
runner="pooling",
dtype=dtype,
max_model_len=None) as vllm_model:
with vllm_runner(
model_info.name, runner="pooling", dtype=dtype, max_model_len=None
) as vllm_model:
assert vllm_model.llm.llm_engine.model_config.is_matryoshka
matryoshka_dimensions = (
vllm_model.llm.llm_engine.model_config.matryoshka_dimensions)
vllm_model.llm.llm_engine.model_config.matryoshka_dimensions
)
assert matryoshka_dimensions is not None
if dimensions not in matryoshka_dimensions:
with pytest.raises(ValueError):
vllm_model.embed(
example_prompts,
pooling_params=PoolingParams(dimensions=dimensions))
example_prompts, pooling_params=PoolingParams(dimensions=dimensions)
)
else:
vllm_outputs = vllm_model.embed(
example_prompts,
pooling_params=PoolingParams(dimensions=dimensions))
example_prompts, pooling_params=PoolingParams(dimensions=dimensions)
)
check_embeddings_close(
embeddings_0_lst=hf_outputs,