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

37
tests/entrypoints/pooling/llm/test_classify.py
View File

@@ -19,12 +19,14 @@ prompts = ["The chef prepared a delicious meal."]
def llm():
# pytest caches the fixture so we use weakref.proxy to
# enable garbage collection
llm = LLM(model=MODEL_NAME,
max_num_batched_tokens=32768,
tensor_parallel_size=1,
gpu_memory_utilization=0.75,
enforce_eager=True,
seed=0)
llm = LLM(
model=MODEL_NAME,
max_num_batched_tokens=32768,
tensor_parallel_size=1,
gpu_memory_utilization=0.75,
enforce_eager=True,
seed=0,
)
yield weakref.proxy(llm)
@@ -35,26 +37,25 @@ def llm():
@pytest.mark.skip_global_cleanup
def test_pooling_params(llm: LLM):
def get_outputs(activation):
outputs = llm.classify(
prompts,
pooling_params=PoolingParams(activation=activation),
use_tqdm=False)
prompts, pooling_params=PoolingParams(activation=activation), use_tqdm=False
)
return torch.tensor([x.outputs.probs for x in outputs])
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(
softmax(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(softmax(wo_activation), w_activation, atol=1e-2), (
"w_activation should be close to activation(wo_activation)."
)
def test_encode_api(llm: LLM):

35
tests/entrypoints/pooling/llm/test_embedding.py
View File

@@ -19,12 +19,14 @@ prompts = ["The chef prepared a delicious meal."]
def llm():
# pytest caches the fixture so we use weakref.proxy to
# enable garbage collection
llm = LLM(model=MODEL_NAME,
max_num_batched_tokens=32768,
tensor_parallel_size=1,
gpu_memory_utilization=0.75,
enforce_eager=True,
seed=0)
llm = LLM(
model=MODEL_NAME,
max_num_batched_tokens=32768,
tensor_parallel_size=1,
gpu_memory_utilization=0.75,
enforce_eager=True,
seed=0,
)
yield weakref.proxy(llm)
@@ -35,21 +37,20 @@ def llm():
@pytest.mark.skip_global_cleanup
def test_pooling_params(llm: LLM):
def get_outputs(normalize):
outputs = llm.embed(prompts,
pooling_params=PoolingParams(normalize=normalize),
use_tqdm=False)
outputs = llm.embed(
prompts, pooling_params=PoolingParams(normalize=normalize), use_tqdm=False
)
return torch.tensor([x.outputs.embedding for x in outputs])
default = get_outputs(normalize=None)
w_normal = get_outputs(normalize=True)
wo_normal = get_outputs(normalize=False)
assert torch.allclose(default, w_normal,
atol=1e-2), "Default should use normal."
assert not torch.allclose(w_normal, wo_normal,
atol=1e-2), "wo_normal should not use normal."
assert torch.allclose(
w_normal, F.normalize(wo_normal, p=2, dim=-1),
atol=1e-2), "w_normal should be close to normal(wo_normal)."
assert torch.allclose(default, w_normal, atol=1e-2), "Default should use normal."
assert not torch.allclose(w_normal, wo_normal, atol=1e-2), (
"wo_normal should not use normal."
)
assert torch.allclose(w_normal, F.normalize(wo_normal, p=2, dim=-1), atol=1e-2), (
"w_normal should be close to normal(wo_normal)."
)

14
tests/entrypoints/pooling/llm/test_encode.py
View File

@@ -31,12 +31,14 @@ TOKEN_IDS = [
def llm():
# pytest caches the fixture so we use weakref.proxy to
# enable garbage collection
llm = LLM(model=MODEL_NAME,
max_num_batched_tokens=32768,
tensor_parallel_size=1,
gpu_memory_utilization=0.75,
enforce_eager=True,
seed=0)
llm = LLM(
model=MODEL_NAME,
max_num_batched_tokens=32768,
tensor_parallel_size=1,
gpu_memory_utilization=0.75,
enforce_eager=True,
seed=0,
)
yield weakref.proxy(llm)

37
tests/entrypoints/pooling/llm/test_reward.py
View File

@@ -19,13 +19,15 @@ prompts = ["The chef prepared a delicious meal."]
def llm():
# pytest caches the fixture so we use weakref.proxy to
# enable garbage collection
llm = LLM(model=MODEL_NAME,
max_num_batched_tokens=32768,
tensor_parallel_size=1,
gpu_memory_utilization=0.75,
enforce_eager=True,
trust_remote_code=True,
seed=0)
llm = LLM(
model=MODEL_NAME,
max_num_batched_tokens=32768,
tensor_parallel_size=1,
gpu_memory_utilization=0.75,
enforce_eager=True,
trust_remote_code=True,
seed=0,
)
yield weakref.proxy(llm)
@@ -36,21 +38,20 @@ def llm():
@pytest.mark.skip_global_cleanup
def test_pooling_params(llm: LLM):
def get_outputs(softmax):
outputs = llm.reward(prompts,
pooling_params=PoolingParams(softmax=softmax),
use_tqdm=False)
outputs = llm.reward(
prompts, pooling_params=PoolingParams(softmax=softmax), use_tqdm=False
)
return torch.cat([x.outputs.data for x in outputs])
default = get_outputs(softmax=None)
w_softmax = get_outputs(softmax=True)
wo_softmax = get_outputs(softmax=False)
assert torch.allclose(default, w_softmax,
atol=1e-2), "Default should use softmax."
assert not torch.allclose(w_softmax, wo_softmax,
atol=1e-2), "wo_softmax should not use softmax."
assert torch.allclose(
softmax(wo_softmax), w_softmax,
atol=1e-2), "w_softmax should be close to softmax(wo_softmax)."
assert torch.allclose(default, w_softmax, atol=1e-2), "Default should use softmax."
assert not torch.allclose(w_softmax, wo_softmax, atol=1e-2), (
"wo_softmax should not use softmax."
)
assert torch.allclose(softmax(wo_softmax), w_softmax, atol=1e-2), (
"w_softmax should be close to softmax(wo_softmax)."
)

35
tests/entrypoints/pooling/llm/test_score.py
View File

@@ -17,12 +17,14 @@ MODEL_NAME = "tomaarsen/Qwen3-Reranker-0.6B-seq-cls"
def llm():
# pytest caches the fixture so we use weakref.proxy to
# enable garbage collection
llm = LLM(model=MODEL_NAME,
max_num_batched_tokens=32768,
tensor_parallel_size=1,
gpu_memory_utilization=0.75,
enforce_eager=True,
seed=0)
llm = LLM(
model=MODEL_NAME,
max_num_batched_tokens=32768,
tensor_parallel_size=1,
gpu_memory_utilization=0.75,
enforce_eager=True,
seed=0,
)
yield weakref.proxy(llm)
@@ -33,7 +35,6 @@ def llm():
@pytest.mark.skip_global_cleanup
def test_pooling_params(llm: LLM):
def get_outputs(activation):
text_1 = "What is the capital of France?"
text_2 = "The capital of France is Paris."
@@ -42,18 +43,20 @@ def test_pooling_params(llm: LLM):
text_1,
text_2,
pooling_params=PoolingParams(activation=activation),
use_tqdm=False)
use_tqdm=False,
)
return torch.tensor([x.outputs.score for x in outputs])
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(
softmax(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(softmax(wo_activation), w_activation, atol=1e-2), (
"w_activation should be close to activation(wo_activation)."
)