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

216
tests/test_config.py
View File

@@ -23,8 +23,8 @@ def test_compile_config_repr_succeeds():
# test that repr(config) succeeds
val = repr(config)
assert 'VllmConfig' in val
assert 'inductor_passes' in val
assert "VllmConfig" in val
assert "inductor_passes" in val
@dataclass
@@ -51,8 +51,7 @@ def test_get_field():
@dataclass
class _TestNestedConfig:
a: _TestConfigFields = field(
default_factory=lambda: _TestConfigFields(a=0))
a: _TestConfigFields = field(default_factory=lambda: _TestConfigFields(a=0))
def test_update_config():
@@ -79,20 +78,19 @@ def test_update_config():
# Can remove once --task option is fully deprecated
@pytest.mark.parametrize(
("model_id", "expected_runner_type", "expected_convert_type",
"expected_task"),
("model_id", "expected_runner_type", "expected_convert_type", "expected_task"),
[
("distilbert/distilgpt2", "generate", "none", "generate"),
("intfloat/multilingual-e5-small", "pooling", "none", "embed"),
("jason9693/Qwen2.5-1.5B-apeach", "pooling", "classify", "classify"),
("cross-encoder/ms-marco-MiniLM-L-6-v2", "pooling", "none",
"classify"),
("cross-encoder/ms-marco-MiniLM-L-6-v2", "pooling", "none", "classify"),
("Qwen/Qwen2.5-Math-RM-72B", "pooling", "none", "reward"),
("openai/whisper-small", "generate", "none", "transcription"),
],
)
def test_auto_task(model_id, expected_runner_type, expected_convert_type,
expected_task):
def test_auto_task(
model_id, expected_runner_type, expected_convert_type, expected_task
):
config = ModelConfig(model_id, task="auto")
assert config.runner_type == expected_runner_type
@@ -101,20 +99,19 @@ def test_auto_task(model_id, expected_runner_type, expected_convert_type,
# Can remove once --task option is fully deprecated
@pytest.mark.parametrize(
("model_id", "expected_runner_type", "expected_convert_type",
"expected_task"),
("model_id", "expected_runner_type", "expected_convert_type", "expected_task"),
[
("distilbert/distilgpt2", "pooling", "embed", "embed"),
("intfloat/multilingual-e5-small", "pooling", "embed", "embed"),
("jason9693/Qwen2.5-1.5B-apeach", "pooling", "classify", "classify"),
("cross-encoder/ms-marco-MiniLM-L-6-v2", "pooling", "classify",
"classify"),
("cross-encoder/ms-marco-MiniLM-L-6-v2", "pooling", "classify", "classify"),
("Qwen/Qwen2.5-Math-RM-72B", "pooling", "embed", "embed"),
("openai/whisper-small", "pooling", "embed", "embed"),
],
)
def test_score_task(model_id, expected_runner_type, expected_convert_type,
expected_task):
def test_score_task(
model_id, expected_runner_type, expected_convert_type, expected_task
):
config = ModelConfig(model_id, task="score")
assert config.runner_type == expected_runner_type
@@ -123,14 +120,14 @@ def test_score_task(model_id, expected_runner_type, expected_convert_type,
# Can remove once --task option is fully deprecated
@pytest.mark.parametrize(
("model_id", "expected_runner_type", "expected_convert_type",
"expected_task"),
("model_id", "expected_runner_type", "expected_convert_type", "expected_task"),
[
("openai/whisper-small", "generate", "none", "transcription"),
],
)
def test_transcription_task(model_id, expected_runner_type,
expected_convert_type, expected_task):
def test_transcription_task(
model_id, expected_runner_type, expected_convert_type, expected_task
):
config = ModelConfig(model_id, task="transcription")
assert config.runner_type == expected_runner_type
@@ -200,8 +197,9 @@ def test_disable_sliding_window(model_id_expected):
assert model_config.max_model_len == expected
@pytest.mark.skipif(current_platform.is_rocm(),
reason="Xformers backend is not supported on ROCm.")
@pytest.mark.skipif(
current_platform.is_rocm(), reason="Xformers backend is not supported on ROCm."
)
def test_get_pooling_config():
model_id = "sentence-transformers/all-MiniLM-L12-v2"
model_config = ModelConfig(model_id)
@@ -211,8 +209,9 @@ def test_get_pooling_config():
assert model_config.pooler_config.pooling_type == PoolingType.MEAN.name
@pytest.mark.skipif(current_platform.is_rocm(),
reason="Xformers backend is not supported on ROCm.")
@pytest.mark.skipif(
current_platform.is_rocm(), reason="Xformers backend is not supported on ROCm."
)
def test_get_pooling_config_from_args():
model_id = "sentence-transformers/all-MiniLM-L12-v2"
pooler_config = PoolerConfig(pooling_type="CLS", normalize=True)
@@ -227,16 +226,18 @@ def test_get_pooling_config_from_args():
("tomaarsen/Qwen3-Reranker-0.6B-seq-cls", "LAST", "LAST"), # LLM
("intfloat/e5-small", "CLS", "MEAN"), # BertModel
("Qwen/Qwen2.5-Math-RM-72B", "ALL", "ALL"), # reward
("Qwen/Qwen2.5-Math-PRM-7B", "STEP", "STEP") # step reward
])
("Qwen/Qwen2.5-Math-PRM-7B", "STEP", "STEP"), # step reward
],
)
def test_default_pooling_type(model_id, default_pooling_type, pooling_type):
model_config = ModelConfig(model_id)
assert model_config._model_info.default_pooling_type == default_pooling_type
assert model_config.pooler_config.pooling_type == pooling_type
@pytest.mark.skipif(current_platform.is_rocm(),
reason="Xformers backend is not supported on ROCm.")
@pytest.mark.skipif(
current_platform.is_rocm(), reason="Xformers backend is not supported on ROCm."
)
def test_get_bert_tokenization_sentence_transformer_config():
model_id = "BAAI/bge-base-en-v1.5"
bge_model_config = ModelConfig(model_id)
@@ -264,17 +265,18 @@ def test_rope_customization():
"rope_theta": TEST_ROPE_THETA,
},
)
assert getattr(llama_model_config.hf_config, "rope_scaling",
None) == TEST_ROPE_SCALING
assert getattr(llama_model_config.hf_config, "rope_theta",
None) == TEST_ROPE_THETA
assert (
getattr(llama_model_config.hf_config, "rope_scaling", None) == TEST_ROPE_SCALING
)
assert getattr(llama_model_config.hf_config, "rope_theta", None) == TEST_ROPE_THETA
assert llama_model_config.max_model_len == 16384
longchat_model_config = ModelConfig("lmsys/longchat-13b-16k")
# Check if LONGCHAT_ROPE_SCALING entries are in longchat_model_config
assert all(
longchat_model_config.hf_config.rope_scaling.get(key) == value
for key, value in LONGCHAT_ROPE_SCALING.items())
for key, value in LONGCHAT_ROPE_SCALING.items()
)
assert longchat_model_config.max_model_len == 16384
longchat_model_config = ModelConfig(
@@ -283,28 +285,37 @@ def test_rope_customization():
"rope_scaling": TEST_ROPE_SCALING,
},
)
assert getattr(longchat_model_config.hf_config, "rope_scaling",
None) == TEST_ROPE_SCALING
assert (
getattr(longchat_model_config.hf_config, "rope_scaling", None)
== TEST_ROPE_SCALING
)
assert longchat_model_config.max_model_len == 4096
@pytest.mark.skipif(current_platform.is_rocm(),
reason="Encoder Decoder models not supported on ROCm.")
@pytest.mark.parametrize(("model_id", "is_encoder_decoder"), [
("facebook/opt-125m", False),
("openai/whisper-tiny", True),
("meta-llama/Llama-3.2-1B-Instruct", False),
])
@pytest.mark.skipif(
current_platform.is_rocm(), reason="Encoder Decoder models not supported on ROCm."
)
@pytest.mark.parametrize(
("model_id", "is_encoder_decoder"),
[
("facebook/opt-125m", False),
("openai/whisper-tiny", True),
("meta-llama/Llama-3.2-1B-Instruct", False),
],
)
def test_is_encoder_decoder(model_id, is_encoder_decoder):
config = ModelConfig(model_id)
assert config.is_encoder_decoder == is_encoder_decoder
@pytest.mark.parametrize(("model_id", "uses_mrope"), [
("facebook/opt-125m", False),
("Qwen/Qwen2-VL-2B-Instruct", True),
])
@pytest.mark.parametrize(
("model_id", "uses_mrope"),
[
("facebook/opt-125m", False),
("Qwen/Qwen2-VL-2B-Instruct", True),
],
)
def test_uses_mrope(model_id, uses_mrope):
config = ModelConfig(model_id)
@@ -338,7 +349,8 @@ def test_generation_config_loading():
model_config = ModelConfig(
model_id,
generation_config="auto",
override_generation_config=override_generation_config)
override_generation_config=override_generation_config,
)
override_result = correct_generation_config.copy()
override_result.update(override_generation_config)
@@ -350,17 +362,19 @@ def test_generation_config_loading():
model_config = ModelConfig(
model_id,
generation_config="vllm",
override_generation_config=override_generation_config)
override_generation_config=override_generation_config,
)
assert model_config.get_diff_sampling_param() == override_generation_config
@pytest.mark.parametrize("pt_load_map_location", [
"cuda",
{
"": "cuda"
},
])
@pytest.mark.parametrize(
"pt_load_map_location",
[
"cuda",
{"": "cuda"},
],
)
def test_load_config_pt_load_map_location(pt_load_map_location):
load_config = LoadConfig(pt_load_map_location=pt_load_map_location)
config = VllmConfig(load_config=load_config)
@@ -369,15 +383,18 @@ def test_load_config_pt_load_map_location(pt_load_map_location):
@pytest.mark.parametrize(
("model_id", "max_model_len", "expected_max_len", "should_raise"), [
("model_id", "max_model_len", "expected_max_len", "should_raise"),
[
("BAAI/bge-reranker-base", None, 512, False),
("BAAI/bge-reranker-base", 256, 256, False),
("BAAI/bge-reranker-base", 513, 512, True),
("deepseek-ai/DeepSeek-R1-Distill-Qwen-7B", None, 131072, False),
("deepseek-ai/DeepSeek-R1-Distill-Qwen-7B", 131073, 131072, True),
])
def test_get_and_verify_max_len(model_id, max_model_len, expected_max_len,
should_raise):
],
)
def test_get_and_verify_max_len(
model_id, max_model_len, expected_max_len, should_raise
):
"""Test get_and_verify_max_len with different configurations."""
model_config = ModelConfig(model_id)
@@ -398,11 +415,14 @@ class MockConfig:
self.model_weights = None
@pytest.mark.parametrize("s3_url", [
"s3://example-bucket-1/model/",
"s3://example-bucket-2/model/",
])
@patch('vllm.transformers_utils.runai_utils.ObjectStorageModel.pull_files')
@pytest.mark.parametrize(
"s3_url",
[
"s3://example-bucket-1/model/",
"s3://example-bucket-2/model/",
],
)
@patch("vllm.transformers_utils.runai_utils.ObjectStorageModel.pull_files")
def test_s3_url_model_tokenizer_paths(mock_pull_files, s3_url):
"""Test that S3 URLs create deterministic local directories for model and
tokenizer."""
@@ -414,22 +434,24 @@ def test_s3_url_model_tokenizer_paths(mock_pull_files, s3_url):
ModelConfig.maybe_pull_model_tokenizer_for_runai(config1, s3_url, s3_url)
# Check that model and tokenizer point to existing directories
assert os.path.exists(
config1.model), f"Model directory does not exist: {config1.model}"
assert os.path.isdir(
config1.model), f"Model path is not a directory: {config1.model}"
assert os.path.exists(
config1.tokenizer
), f"Tokenizer directory does not exist: {config1.tokenizer}"
assert os.path.isdir(
config1.tokenizer
), f"Tokenizer path is not a directory: {config1.tokenizer}"
assert os.path.exists(config1.model), (
f"Model directory does not exist: {config1.model}"
)
assert os.path.isdir(config1.model), (
f"Model path is not a directory: {config1.model}"
)
assert os.path.exists(config1.tokenizer), (
f"Tokenizer directory does not exist: {config1.tokenizer}"
)
assert os.path.isdir(config1.tokenizer), (
f"Tokenizer path is not a directory: {config1.tokenizer}"
)
# Verify that the paths are different from the original S3 URL
assert config1.model != s3_url, (
"Model path should be converted to local directory")
assert config1.model != s3_url, "Model path should be converted to local directory"
assert config1.tokenizer != s3_url, (
"Tokenizer path should be converted to local directory")
"Tokenizer path should be converted to local directory"
)
# Store the original paths
created_model_dir = config1.model
@@ -440,27 +462,31 @@ def test_s3_url_model_tokenizer_paths(mock_pull_files, s3_url):
ModelConfig.maybe_pull_model_tokenizer_for_runai(config2, s3_url, s3_url)
# Check that the new directories exist
assert os.path.exists(
config2.model), f"Model directory does not exist: {config2.model}"
assert os.path.isdir(
config2.model), f"Model path is not a directory: {config2.model}"
assert os.path.exists(
config2.tokenizer
), f"Tokenizer directory does not exist: {config2.tokenizer}"
assert os.path.isdir(
config2.tokenizer
), f"Tokenizer path is not a directory: {config2.tokenizer}"
assert os.path.exists(config2.model), (
f"Model directory does not exist: {config2.model}"
)
assert os.path.isdir(config2.model), (
f"Model path is not a directory: {config2.model}"
)
assert os.path.exists(config2.tokenizer), (
f"Tokenizer directory does not exist: {config2.tokenizer}"
)
assert os.path.isdir(config2.tokenizer), (
f"Tokenizer path is not a directory: {config2.tokenizer}"
)
# Verify that the paths are deterministic (same as before)
assert config2.model == created_model_dir, (
f"Model paths are not deterministic. "
f"Original: {created_model_dir}, New: {config2.model}")
f"Original: {created_model_dir}, New: {config2.model}"
)
assert config2.tokenizer == create_tokenizer_dir, (
f"Tokenizer paths are not deterministic. "
f"Original: {create_tokenizer_dir}, New: {config2.tokenizer}")
f"Original: {create_tokenizer_dir}, New: {config2.tokenizer}"
)
@patch('vllm.transformers_utils.runai_utils.ObjectStorageModel.pull_files')
@patch("vllm.transformers_utils.runai_utils.ObjectStorageModel.pull_files")
def test_s3_url_different_models_create_different_directories(mock_pull_files):
"""Test that different S3 URLs create different local directories."""
# Mock pull_files to avoid actually downloading files during tests
@@ -479,16 +505,16 @@ def test_s3_url_different_models_create_different_directories(mock_pull_files):
# Verify that different URLs produce different directories
assert config1.model != config2.model, (
f"Different S3 URLs should create different model directories. "
f"URL1 model: {config1.model}, URL2 model: {config2.model}")
f"URL1 model: {config1.model}, URL2 model: {config2.model}"
)
assert config1.tokenizer != config2.tokenizer, (
f"Different S3 URLs should create different tokenizer directories. "
f"URL1 tokenizer: {config1.tokenizer}, "
f"URL2 tokenizer: {config2.tokenizer}")
f"URL2 tokenizer: {config2.tokenizer}"
)
# Verify that both sets of directories exist
assert os.path.exists(config1.model) and os.path.isdir(config1.model)
assert os.path.exists(config1.tokenizer) and os.path.isdir(
config1.tokenizer)
assert os.path.exists(config1.tokenizer) and os.path.isdir(config1.tokenizer)
assert os.path.exists(config2.model) and os.path.isdir(config2.model)
assert os.path.exists(config2.tokenizer) and os.path.isdir(
config2.tokenizer)
assert os.path.exists(config2.tokenizer) and os.path.isdir(config2.tokenizer)