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

186
tests/entrypoints/pooling/openai/test_score.py
View File

@@ -12,14 +12,8 @@ from tests.utils import RemoteOpenAIServer
from vllm.entrypoints.openai.protocol import ScoreResponse
MODELS = [
{
"name": "BAAI/bge-reranker-v2-m3",
"is_cross_encoder": True
},
{
"name": "BAAI/bge-base-en-v1.5",
"is_cross_encoder": False
},
{"name": "BAAI/bge-reranker-v2-m3", "is_cross_encoder": True},
{"name": "BAAI/bge-base-en-v1.5", "is_cross_encoder": False},
]
DTYPE = "half"
@@ -28,9 +22,7 @@ def run_transformers(hf_model, model, text_pairs):
if model["is_cross_encoder"]:
return hf_model.predict(text_pairs).tolist()
else:
hf_embeddings = [
hf_model.encode(text_pair) for text_pair in text_pairs
]
hf_embeddings = [hf_model.encode(text_pair) for text_pair in text_pairs]
return [
F.cosine_similarity(tensor(pair[0]), tensor(pair[1]), dim=0)
for pair in hf_embeddings
@@ -54,8 +46,9 @@ def server(model: dict[str, Any]):
def runner(model: dict[str, Any], hf_runner):
kwargs = {
"dtype": DTYPE,
"is_cross_encoder" if model["is_cross_encoder"]\
else "is_sentence_transformer": True
"is_cross_encoder"
if model["is_cross_encoder"]
else "is_sentence_transformer": True,
}
with hf_runner(model["name"], **kwargs) as hf_model:
@@ -63,21 +56,23 @@ def runner(model: dict[str, Any], hf_runner):
class TestModel:
def test_text_1_str_text_2_list(self, server: RemoteOpenAIServer,
model: dict[str, Any], runner):
def test_text_1_str_text_2_list(
self, server: RemoteOpenAIServer, model: dict[str, Any], runner
):
text_1 = "What is the capital of France?"
text_2 = [
"The capital of Brazil is Brasilia.",
"The capital of France is Paris."
"The capital of France is Paris.",
]
score_response = requests.post(server.url_for("score"),
json={
"model": model["name"],
"text_1": text_1,
"text_2": text_2,
})
score_response = requests.post(
server.url_for("score"),
json={
"model": model["name"],
"text_1": text_1,
"text_2": text_2,
},
)
score_response.raise_for_status()
score = ScoreResponse.model_validate(score_response.json())
@@ -93,23 +88,26 @@ class TestModel:
for i in range(len(vllm_outputs)):
assert hf_outputs[i] == pytest.approx(vllm_outputs[i], rel=0.01)
def test_text_1_list_text_2_list(self, server: RemoteOpenAIServer,
model: dict[str, Any], runner):
def test_text_1_list_text_2_list(
self, server: RemoteOpenAIServer, model: dict[str, Any], runner
):
text_1 = [
"What is the capital of the United States?",
"What is the capital of France?"
"What is the capital of France?",
]
text_2 = [
"The capital of Brazil is Brasilia.",
"The capital of France is Paris."
"The capital of France is Paris.",
]
score_response = requests.post(server.url_for("score"),
json={
"model": model["name"],
"text_1": text_1,
"text_2": text_2,
})
score_response = requests.post(
server.url_for("score"),
json={
"model": model["name"],
"text_1": text_1,
"text_2": text_2,
},
)
score_response.raise_for_status()
score = ScoreResponse.model_validate(score_response.json())
@@ -125,17 +123,20 @@ class TestModel:
for i in range(len(vllm_outputs)):
assert hf_outputs[i] == pytest.approx(vllm_outputs[i], rel=0.01)
def test_text_1_str_text_2_str(self, server: RemoteOpenAIServer,
model: dict[str, Any], runner):
def test_text_1_str_text_2_str(
self, server: RemoteOpenAIServer, model: dict[str, Any], runner
):
text_1 = "What is the capital of France?"
text_2 = "The capital of France is Paris."
score_response = requests.post(server.url_for("score"),
json={
"model": model["name"],
"text_1": text_1,
"text_2": text_2,
})
score_response = requests.post(
server.url_for("score"),
json={
"model": model["name"],
"text_1": text_1,
"text_2": text_2,
},
)
score_response.raise_for_status()
score = ScoreResponse.model_validate(score_response.json())
@@ -151,40 +152,41 @@ class TestModel:
for i in range(len(vllm_outputs)):
assert hf_outputs[i] == pytest.approx(vllm_outputs[i], rel=0.01)
def test_score_max_model_len(self, server: RemoteOpenAIServer,
model: dict[str, Any]):
def test_score_max_model_len(
self, server: RemoteOpenAIServer, model: dict[str, Any]
):
text_1 = "What is the capital of France?" * 20
text_2 = [
"The capital of Brazil is Brasilia.",
"The capital of France is Paris."
"The capital of France is Paris.",
]
score_response = requests.post(server.url_for("score"),
json={
"model": model["name"],
"text_1": text_1,
"text_2": text_2,
})
score_response = requests.post(
server.url_for("score"),
json={
"model": model["name"],
"text_1": text_1,
"text_2": text_2,
},
)
assert score_response.status_code == 400
# Assert just a small fragments of the response
assert "Please reduce the length of the input." in \
score_response.text
assert "Please reduce the length of the input." in score_response.text
# Test truncation
score_response = requests.post(server.url_for("score"),
json={
"model": model["name"],
"text_1": text_1,
"text_2": text_2,
"truncate_prompt_tokens": 101
})
score_response = requests.post(
server.url_for("score"),
json={
"model": model["name"],
"text_1": text_1,
"text_2": text_2,
"truncate_prompt_tokens": 101,
},
)
assert score_response.status_code == 400
assert "Please, select a smaller truncation size." in \
score_response.text
assert "Please, select a smaller truncation size." in score_response.text
def test_invocations(self, server: RemoteOpenAIServer, model: dict[str,
Any]):
def test_invocations(self, server: RemoteOpenAIServer, model: dict[str, Any]):
text_1 = "What is the capital of France?"
text_2 = "The capital of France is Paris."
@@ -194,59 +196,61 @@ class TestModel:
"text_2": text_2,
}
score_response = requests.post(server.url_for("score"),
json=request_args)
score_response = requests.post(server.url_for("score"), json=request_args)
score_response.raise_for_status()
invocation_response = requests.post(server.url_for("invocations"),
json=request_args)
invocation_response = requests.post(
server.url_for("invocations"), json=request_args
)
invocation_response.raise_for_status()
score_output = score_response.json()
invocation_output = invocation_response.json()
assert score_output.keys() == invocation_output.keys()
for score_data, invocation_data in zip(score_output["data"],
invocation_output["data"]):
for score_data, invocation_data in zip(
score_output["data"], invocation_output["data"]
):
assert score_data.keys() == invocation_data.keys()
assert score_data["score"] == pytest.approx(
invocation_data["score"], rel=0.05)
invocation_data["score"], rel=0.05
)
# TODO: reset this tolerance to 0.01 once we find
# an alternative to flash_attn with bfloat16
def test_activation(self, server: RemoteOpenAIServer, model: dict[str,
Any]):
def test_activation(self, server: RemoteOpenAIServer, model: dict[str, Any]):
def get_outputs(activation):
text_1 = "What is the capital of France?"
text_2 = "The capital of France is Paris."
response = requests.post(server.url_for("score"),
json={
"model": model["name"],
"text_1": text_1,
"text_2": text_2,
"activation": activation
})
response = requests.post(
server.url_for("score"),
json={
"model": model["name"],
"text_1": text_1,
"text_2": text_2,
"activation": activation,
},
)
if response.status_code != 200:
return response
outputs = response.json()
return torch.tensor([x['score'] for x in outputs["data"]])
return torch.tensor([x["score"] for x in outputs["data"]])
if model["is_cross_encoder"]:
default = get_outputs(activation=None)
w_activation = get_outputs(activation=True)
wo_activation = get_outputs(activation=False)
assert torch.allclose(default, w_activation,
atol=1e-2), "Default should use activation."
assert not torch.allclose(
w_activation, wo_activation,
atol=1e-2), "wo_activation should not use activation."
assert torch.allclose(
F.sigmoid(wo_activation), w_activation, atol=1e-2
), "w_activation should be close to activation(wo_activation)."
assert torch.allclose(default, w_activation, atol=1e-2), (
"Default should use activation."
)
assert not torch.allclose(w_activation, wo_activation, atol=1e-2), (
"wo_activation should not use activation."
)
assert torch.allclose(F.sigmoid(wo_activation), w_activation, atol=1e-2), (
"w_activation should be close to activation(wo_activation)."
)
else:
get_outputs(activation=None)