iHore/vLLM
8
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

[Refactor] Remove dead tests code (#58916)

Browse Source 
Signed-off-by: yewentao256 <zhyanwentao@126.com>
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
This commit is contained in:
Wentao Ye
2026-09-30 04:39:03 +00:00
committed by GitHub
co-authored by mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
parent 90e13fc757
commit 834ad456d3
17 changed files with 3 additions and 488 deletions
Download Patch File Download Diff File Expand all files Collapse all files
.buildkite/test-amd.yaml
+1 -1
View File
@@ -586,7 +586,7 @@ steps:
- tests/utils.py
- tests/models/test_vision.py
commands:
- case "$$BUILDKITE_PARALLEL_JOB" in 0) TARGET_TEST_SUITE=MI300 pytest basic_correctness/ -v -s -m 'distributed(num_gpus=2)' && HIP_VISIBLE_DEVICES=0,1 pytest -v -s model_executor/model_loader/test_sharded_state_loader.py -m '(not slow_test)' ;; 1) pytest models/transformers/test_backend.py -v -s -m 'distributed(num_gpus=2)' && pytest models/language -v -s -m 'distributed(num_gpus=2)' && VLLM_ROCM_USE_AITER=1 pytest -v -s models/test_hyv4_rocm.py -m 'distributed(num_gpus=2)' ;; 2) pytest models/multimodal -v -s -m 'distributed(num_gpus=2)' --ignore models/multimodal/generation/test_whisper.py --ignore models/multimodal/generation/test_phi4siglip.py && pytest models/multimodal/generation/test_phi4siglip.py -v -s -m 'distributed(num_gpus=2)' && VLLM_WORKER_MULTIPROC_METHOD=spawn pytest models/multimodal/generation/test_whisper.py -v -s -m 'distributed(num_gpus=2)' && pytest -v -s models/test_vision.py -m 'distributed(num_gpus=2)' ;; *) exit 2 ;; esac
- case "$$BUILDKITE_PARALLEL_JOB" in 0) TARGET_TEST_SUITE=MI300 pytest basic_correctness/ -v -s -m 'distributed(num_gpus=2)' && HIP_VISIBLE_DEVICES=0,1 pytest -v -s model_executor/model_loader/test_sharded_state_loader.py -m '(not slow_test)' ;; 1) pytest models/transformers/test_backend.py -v -s -m 'distributed(num_gpus=2)' && pytest models/language -v -s -m 'distributed(num_gpus=2)' && VLLM_ROCM_USE_AITER=1 pytest -v -s models/test_hyv4_rocm.py -m 'distributed(num_gpus=2)' ;; 2) pytest models/multimodal -v -s -m 'distributed(num_gpus=2)' --ignore models/multimodal/generation/test_whisper.py && VLLM_WORKER_MULTIPROC_METHOD=spawn pytest models/multimodal/generation/test_whisper.py -v -s -m 'distributed(num_gpus=2)' && pytest -v -s models/test_vision.py -m 'distributed(num_gpus=2)' ;; *) exit 2 ;; esac
#---------------------------------------------------------- mi300 · compile ----------------------------------------------------------#
.buildkite/test_areas/models_distributed.yaml
+2 -4
View File
@@ -26,8 +26,7 @@ steps:
pytest models/language -v -s -m 'distributed(num_gpus=2)'
;;
2)
pytest models/multimodal/generation/test_phi4siglip.py -v -s -m 'distributed(num_gpus=2)'
pytest models/multimodal -v -s -m 'distributed(num_gpus=2)' --ignore models/multimodal/generation/test_whisper.py --ignore models/multimodal/generation/test_phi4siglip.py
pytest models/multimodal -v -s -m 'distributed(num_gpus=2)' --ignore models/multimodal/generation/test_whisper.py
VLLM_WORKER_MULTIPROC_METHOD=spawn pytest models/multimodal/generation/test_whisper.py -v -s -m 'distributed(num_gpus=2)'
;;
*) exit 2 ;;
@@ -72,8 +71,7 @@ steps:
VLLM_ROCM_USE_AITER=1 pytest -v -s models/test_hyv4_rocm.py -m 'distributed(num_gpus=2)'
;;
2)
pytest models/multimodal -v -s -m 'distributed(num_gpus=2)' --ignore models/multimodal/generation/test_whisper.py --ignore models/multimodal/generation/test_phi4siglip.py &&
pytest models/multimodal/generation/test_phi4siglip.py -v -s -m 'distributed(num_gpus=2)' &&
pytest models/multimodal -v -s -m 'distributed(num_gpus=2)' --ignore models/multimodal/generation/test_whisper.py &&
VLLM_WORKER_MULTIPROC_METHOD=spawn pytest models/multimodal/generation/test_whisper.py -v -s -m 'distributed(num_gpus=2)'
;;
*) exit 2 ;;
tests/compile/passes/test_fusion.py
-4
View File
@@ -60,10 +60,6 @@ from vllm.utils.deep_gemm import (
is_deep_gemm_supported,
)
FP8_DTYPE = current_platform.fp8_dtype()
RMS_ADD_OP = torch.ops._C.fused_add_rms_norm.default
# Kernel and group_shape combinations: (kernel, group_shape)
# CUDA kernels
CUDA_KERNEL_GROUPSHAPE_COMBINATIONS = [
tests/compile/passes/test_fusion_attn.py
-1
View File
@@ -248,7 +248,6 @@ class TestAttentionNvfp4QuantPatternModel(AttentionQuantPatternModel):
PATTERN_TEST_MODELS_FP8: list[tuple[str, type]] = []
PATTERN_TEST_MODELS_FP4: list[tuple[str, type]] = []
HEADS: list[tuple[int, int]] = []
SPLIT_ATTENTION: list[bool] = []
BACKENDS_FP8: list[AttentionBackendEnum] = []
BACKENDS_FP4: list[AttentionBackendEnum] = []
tests/conftest.py
-5
View File
@@ -317,11 +317,6 @@ def example_system_message() -> str:
return f.read()
@pytest.fixture
def example_long_prompts() -> list[str]:
return [prompt for filename in _LONG_PROMPTS for prompt in _read_prompts(filename)]
@pytest.fixture(scope="session")
def image_assets() -> ImageTestAssets:
return IMAGE_ASSETS
tests/entrypoints/conftest.py
-120
View File
@@ -14,16 +14,6 @@ def sample_prompts():
]
@pytest.fixture
def sample_token_ids():
return [
[0],
[0, 1],
[0, 2, 1],
[0, 3, 1, 2],
]
@pytest.fixture
def sample_regex():
return (
@@ -61,104 +51,6 @@ def sample_json_schema():
}
@pytest.fixture
def sample_complex_json_schema():
return {
"type": "object",
"properties": {
"score": {
"type": "integer",
"minimum": 0,
"maximum": 100, # Numeric range
},
"grade": {
"type": "string",
"pattern": "^[A-D]$", # Regex pattern
},
"email": {
"type": "string",
"pattern": "^[a-zA-Z0-9._%+-]+@[a-zA-Z0-9.-]+\\.[a-zA-Z]{2,}$",
},
"tags": {
"type": "array",
"items": {
"type": "string",
# Combining length and pattern restrictions
"pattern": "^[a-z]{1,10}$",
},
},
},
"required": ["score", "grade", "email", "tags"],
}
@pytest.fixture
def sample_definition_json_schema():
return {
"$defs": {
"Step": {
"properties": {
"explanation": {"title": "Explanation", "type": "string"},
"output": {"title": "Output", "type": "string"},
},
"required": ["explanation", "output"],
"title": "Step",
"type": "object",
}
},
"properties": {
"steps": {
"items": {"$ref": "#/$defs/Step"},
"title": "Steps",
"type": "array",
},
"final_answer": {"title": "Final Answer", "type": "string"},
},
"required": ["steps", "final_answer"],
"title": "MathReasoning",
"type": "object",
}
@pytest.fixture
def sample_enum_json_schema():
return {
"type": "object",
"properties": {
"status": {
"type": "string",
"enum": ["active", "inactive", "pending"], # Literal values using enum
},
"priority": {
"type": "string",
"enum": ["low", "medium", "high", "critical"],
},
"category": {
"type": "object",
"properties": {
"type": {
"type": "string",
"enum": ["bug", "feature", "improvement"],
},
"severity": {
"type": "integer",
"enum": [1, 2, 3, 4, 5], # Enum can also contain numbers
},
},
"required": ["type", "severity"],
},
"flags": {
"type": "array",
"items": {
"type": "string",
"enum": ["urgent", "blocked", "needs_review", "approved"],
},
},
},
"required": ["status", "priority", "category", "flags"],
}
@pytest.fixture
def sample_structured_outputs_choices():
return [
@@ -175,18 +67,6 @@ def sample_structured_outputs_choices():
]
@pytest.fixture
def sample_sql_statements():
return """
start: select_statement
select_statement: "SELECT" column "from" table "where" condition
column: "col_1" | "col_2"
table: "table_1" | "table_2"
condition: column "=" number
number: "1" | "2"
"""
@pytest.fixture(scope="session")
def qwen3_lora_files():
"""Download Qwen3 LoRA files once per test session."""
tests/entrypoints/launchers/test_dp_supervisor.py
-1
View File
@@ -47,7 +47,6 @@ _SUPERVISOR_PORT = 19256
_CHILD_PORT_BASE = 18000
_N_CHILDREN = 2
_PROBE_INTERVAL = 1.0
_POLL_INTERVAL = 1.0
# ---------------------------------------------------------------------------
tests/entrypoints/openai/responses/test_serving_responses.py
-8
View File
@@ -127,14 +127,6 @@ def test_serialize_message_pydantic_model_returns_dict() -> None:
assert serialized["message"] == "hello"
@pytest.fixture
def mock_serving_responses():
"""Create a mock OpenAIServingResponses instance."""
serving_responses = MagicMock(spec=OpenAIServingResponses)
serving_responses.tool_server = MagicMock(spec=ToolServer)
return serving_responses
@pytest.fixture
def mock_context():
"""Create a mock conversation context."""
tests/entrypoints/tool_parsers/test_openai_tool_parser.py
-3
View File
@@ -117,10 +117,7 @@ MESSAGES_INVALID_CALL = [
# Expected outputs
FUNC_CALC = "calculator"
FUNC_ARGS_CALC = '{"expression":"123 + 456"}'
FUNC_TIME = "get_time"
FUNC_ARGS_TIME = '{"city": "New York"}'
# ==========================================================
tests/kernels/attention/test_attention.py
-1
View File
@@ -26,7 +26,6 @@ NUM_BLOCKS = 4321 # Arbitrary values for testing
PARTITION_SIZE_ROCM = 256
DTYPES = [torch.bfloat16, torch.float16]
NUM_GEN_SEQS = [7] # Arbitrary values for testing
NUM_PREFILL_SEQS = [3] # Arbitrary values for testing
NUM_HEADS = [(32, 8), (40, 40), (64, 8)] # Arbitrary values for testing
# Head sizes supported by the ROCm paged attention kernel.
tests/kernels/moe/utils.py
-92
View File
@@ -7,7 +7,6 @@ import torch
import vllm._custom_ops as ops
from tests.kernels.quant_utils import per_block_cast_to_int8
from tests.kernels.quantization.nvfp4_utils import FLOAT4_E2M1_MAX, FLOAT8_E4M3_MAX
from vllm.model_executor.layers.activation import SiluAndMul
from vllm.model_executor.layers.fused_moe.activation import MoEActivation
from vllm.model_executor.layers.fused_moe.all2all_utils import (
maybe_make_prepare_finalize,
@@ -552,97 +551,6 @@ class TestMLP(torch.nn.Module):
return x
class RealMLP(torch.nn.Module):
def __init__(
self,
hidden_size: int,
intermediate_size: int,
w1: torch.Tensor,
w2: torch.Tensor,
hidden_act: str = "silu",
quant_config=None,
reduce_results: bool = True,
prefix: str = "",
w1_s: torch.Tensor | None = None,
w2_s: torch.Tensor | None = None,
) -> None:
from vllm.model_executor.layers.linear import (
MergedColumnParallelLinear,
RowParallelLinear,
)
super().__init__()
self.gate_up_proj = MergedColumnParallelLinear(
hidden_size,
[intermediate_size] * 2,
bias=False,
quant_config=quant_config,
prefix=f"{prefix}.gate_up_proj",
)
self.gate_up_proj.register_parameter(
"weight", torch.nn.Parameter(w1, requires_grad=False)
)
self.gate_up_proj.register_parameter(
"weight_scale", torch.nn.Parameter(w1_s, requires_grad=False)
)
self.gate_up_proj.register_parameter(
"input_scale", None
) # torch.nn.Parameter(None, requires_grad=False))
self.down_proj = RowParallelLinear(
intermediate_size,
hidden_size,
bias=False,
quant_config=quant_config,
reduce_results=reduce_results,
prefix=f"{prefix}.down_proj",
)
self.down_proj.register_parameter(
"weight", torch.nn.Parameter(w2, requires_grad=False)
)
self.down_proj.register_parameter(
"weight_scale", torch.nn.Parameter(w2_s, requires_grad=False)
)
self.down_proj.register_parameter(
"input_scale", None
) # torch.nn.Parameter(None, requires_grad=False))
if hidden_act != "silu":
raise ValueError(
f"Unsupported activation: {hidden_act}. Only silu is supported for now."
)
self.act_fn = SiluAndMul()
def forward(self, x):
gate_up, _ = self.gate_up_proj(x)
x = self.act_fn(gate_up)
x, _ = self.down_proj(x)
return x
def make_shared_experts_with_weights(
N: int,
K: int,
in_dtype: torch.dtype,
w1: torch.Tensor,
w2: torch.Tensor,
w1_s: torch.Tensor | None = None,
w2_s: torch.Tensor | None = None,
quant_dtype: torch.dtype | str | None = None,
) -> torch.nn.Module:
old_dtype = torch.get_default_dtype()
try:
torch.set_default_dtype(in_dtype)
if quant_dtype == torch.float8_e4m3fn:
from vllm.model_executor.layers.quantization.fp8 import Fp8Config
quant_config = Fp8Config()
else:
quant_config = None
return RealMLP(K, N, w1, w2, "silu", quant_config, w1_s=w1_s, w2_s=w2_s)
finally:
torch.set_default_dtype(old_dtype)
def modular_triton_fused_moe(
moe_config: FusedMoEConfig,
quant_config: FusedMoEQuantConfig,
tests/lora/conftest.py
-18
View File
@@ -76,18 +76,6 @@ def dist_init():
cleanup_dist_env_and_memory(shutdown_ray=True)
@pytest.fixture
def dist_init_torch_only():
if torch.distributed.is_initialized():
return
backend = current_platform.dist_backend
temp_file = tempfile.mkstemp()[1]
torch.distributed.init_process_group(
world_size=1, rank=0, init_method=f"file://{temp_file}", backend=backend
)
class DummyLoRAModel(nn.Sequential, SupportsLoRA):
pass
@@ -208,12 +196,6 @@ def qwen2vl_lora_files():
return hf_api().snapshot_download(repo_id="jeeejeee/qwen2-vl-lora-pokemon")
@pytest.fixture(scope="session")
def qwen25vl_base_huggingface_id():
# used as a base model for testing with qwen25vl lora adapter
return "Qwen/Qwen2.5-VL-3B-Instruct"
@pytest.fixture(scope="session")
def qwen25vl_lora_files():
return hf_api().snapshot_download(repo_id="jeeejeee/qwen25-vl-lora-pokemon")
tests/models/multimodal/generation/test_phi4siglip.py
-198
View File
@@ -1,198 +0,0 @@
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
from collections.abc import Sequence
from importlib.metadata import version
import pytest
import regex as re
from packaging.version import Version
from transformers import AutoModelForCausalLM, AutoTokenizer
from vllm.logprobs import SampleLogprobs
from vllm.multimodal.image import rescale_image_size
from ....conftest import (
IMAGE_ASSETS,
HfRunner,
PromptImageInput,
VllmRunner,
)
from ....utils import multi_gpu_test
from ...utils import check_logprobs_close
pytestmark = pytest.mark.skipif(
Version("5.0") <= Version(version("transformers")),
reason=(
"vllm upgraded transformers above v5.4 where HF model custom code uses siglip2 "
"internals (filter_out_non_signature_kwargs) removed by "
"huggingface/transformers#43514"
),
)
MODEL_ID = "microsoft/Phi-4-reasoning-vision-15B"
HF_IMAGE_PROMPTS = IMAGE_ASSETS.prompts(
{
"stop_sign": "<|user|>\n<image>\nWhat's the content of the image?<|end|>\n<|assistant|>\n", # noqa: E501
"cherry_blossom": "<|user|>\n<image>\nPlease infer the season with reason in details.<|end|>\n<|assistant|>\n", # noqa: E501
}
)
HF_MULTIIMAGE_IMAGE_PROMPT = (
"<|user|>\n<image>\n<image>\nDescribe these images.<|end|>\n<|assistant|>\n" # noqa: E501
)
DTYPE = "half"
MAX_TOKENS = 128
NUM_LOGPROBS = 10
def vllm_to_hf_output(
vllm_output: tuple[list[int], str, SampleLogprobs | None], model: str
):
"""Sanitize vllm output to be comparable with hf output."""
_, output_str, out_logprobs = vllm_output
output_str_without_image = re.sub(r"(<image>)+", "", output_str)
if output_str_without_image and output_str_without_image[0] == " ":
output_str_without_image = output_str_without_image[1:]
hf_output_str = output_str_without_image + "<|end|><|endoftext|>"
tokenizer = AutoTokenizer.from_pretrained(model, trust_remote_code=True)
hf_output_ids = tokenizer.encode(output_str_without_image)
if hf_output_ids and hf_output_ids[0] == tokenizer.bos_token_id:
hf_output_ids = hf_output_ids[1:]
return hf_output_ids, hf_output_str, out_logprobs
def _build_single_image_inputs(
image_assets,
) -> list[tuple[list[str], PromptImageInput]]:
"""Build single-image inputs for all size_factors at once."""
images = [asset.pil_image for asset in image_assets]
all_inputs: list[tuple[list[str], PromptImageInput]] = []
for size_factors in [[1.0], [0.25, 0.5, 1.0]]:
for image, prompt in zip(images, HF_IMAGE_PROMPTS):
all_inputs.append(
(
[prompt for _ in size_factors],
[rescale_image_size(image, f) for f in size_factors],
)
)
return all_inputs
def _build_multi_image_inputs(
image_assets,
) -> list[tuple[list[str], PromptImageInput]]:
"""Build multi-image inputs for all size_factors at once."""
images = [asset.pil_image for asset in image_assets]
all_inputs: list[tuple[list[str], PromptImageInput]] = []
for size_factors in [[0.5], [0.15, 0.30]]:
all_inputs.append(
(
[HF_MULTIIMAGE_IMAGE_PROMPT for _ in size_factors],
[
[rescale_image_size(image, factor) for image in images]
for factor in size_factors
],
)
)
return all_inputs
def _run_and_compare(
hf_runner: type[HfRunner],
vllm_runner: type[VllmRunner],
all_inputs: Sequence[tuple[list[str], PromptImageInput]],
model: str,
max_model_len: int,
max_num_seqs: int,
mm_limit: int,
gpu_memory_utilization: float,
):
"""Load each runner once, run all inputs, then compare."""
# NOTE: run vLLM first, then HF. vLLM needs a fresh process without
# cuda initialization; running HF first would break the multiprocessing
# backend with fork method.
with vllm_runner(
model,
runner="generate",
max_model_len=max_model_len,
max_num_seqs=max_num_seqs,
gpu_memory_utilization=gpu_memory_utilization,
dtype=DTYPE,
limit_mm_per_prompt={"image": mm_limit},
tensor_parallel_size=2,
trust_remote_code=True,
enforce_eager=True,
) as vllm_model:
vllm_outputs_per_case = [
vllm_model.generate_greedy_logprobs(
prompts,
MAX_TOKENS,
num_logprobs=NUM_LOGPROBS,
images=images,
)
for prompts, images in all_inputs
]
hf_model_kwargs = {"_attn_implementation": "sdpa", "device_map": "auto"}
with hf_runner(
model,
dtype=DTYPE,
model_kwargs=hf_model_kwargs,
auto_cls=AutoModelForCausalLM,
trust_remote_code=True,
) as hf_model:
hf_outputs_per_case = [
hf_model.generate_greedy_logprobs_limit(
prompts,
MAX_TOKENS,
num_logprobs=NUM_LOGPROBS,
images=images,
)
for prompts, images in all_inputs
]
for hf_outputs, vllm_outputs in zip(hf_outputs_per_case, vllm_outputs_per_case):
check_logprobs_close(
outputs_0_lst=hf_outputs,
outputs_1_lst=vllm_outputs,
name_0="hf",
name_1="vllm",
)
@multi_gpu_test(num_gpus=2)
@pytest.mark.parametrize("model", [MODEL_ID])
def test_models(hf_runner, vllm_runner, image_assets, model) -> None:
all_inputs = _build_single_image_inputs(image_assets)
_run_and_compare(
hf_runner,
vllm_runner,
all_inputs,
model,
max_model_len=8192,
max_num_seqs=2,
mm_limit=1,
gpu_memory_utilization=0.80,
)
@multi_gpu_test(num_gpus=2)
@pytest.mark.parametrize("model", [MODEL_ID])
def test_multi_images_models(hf_runner, vllm_runner, image_assets, model) -> None:
all_inputs = _build_multi_image_inputs(image_assets)
_run_and_compare(
hf_runner,
vllm_runner,
all_inputs,
model,
max_model_len=8192,
max_num_seqs=2,
mm_limit=2,
gpu_memory_utilization=0.80,
)
tests/reasoning/test_kimi_k2_reasoning_parser.py
-15
View File
@@ -1,7 +1,6 @@
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
from unittest.mock import MagicMock
import pytest
@@ -13,20 +12,6 @@ from vllm.tokenizers import get_tokenizer
REASONING_MODEL_NAME = "moonshotai/Kimi-K2.5"
@pytest.fixture
def mock_kimi_k2_tokenizer():
tokenizer = MagicMock()
tokenizer.get_vocab.return_value = {
"<think>": 100,
"</think>": 101,
"<|tool_calls_section_begin|>": 200,
"<|tool_calls_section_end|>": 201,
"<|tool_call_begin|>": 202,
"<|tool_call_end|>": 203,
}
return tokenizer
@pytest.fixture(scope="module")
def kimi_k2_tokenizer():
return get_tokenizer(tokenizer_name=REASONING_MODEL_NAME, trust_remote_code=True)
tests/tracing/conftest.py
-6
View File
@@ -142,9 +142,3 @@ def trace_service() -> Generator[FakeTraceService, None, None]:
yield service
server.stop(grace=None)
@pytest.fixture
def trace_server_address() -> str:
"""Returns the address of the fake trace server."""
return FAKE_TRACE_SERVER_ADDRESS
tests/v1/engine/conftest.py
-4
View File
@@ -2,7 +2,6 @@
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
import pytest
import torch
from transformers import AutoTokenizer
from tests.v1.engine.utils import (
@@ -19,9 +18,6 @@ from vllm.engine.arg_utils import EngineArgs
from ...distributed.conftest import publisher_config, random_port # noqa: F401
EngineCoreSampleLogprobsType = list[tuple[torch.Tensor, torch.Tensor]]
EngineCorePromptLogprobsType = tuple[torch.Tensor, torch.Tensor]
def _build_test_vectors_no_logprobs() -> DummyOutputProcessorTestVectors:
"""Generate output processor dummy test vectors, without logprobs.
tests/v1/engine/test_llm_engine.py
-7
View File
@@ -49,13 +49,6 @@ def vllm_model(vllm_runner, request):
yield vllm_model
@pytest.fixture(scope="function")
def vllm_model_apc(vllm_runner):
"""VllmRunner test fixture with APC."""
with _vllm_model(True, vllm_runner) as vllm_model:
yield vllm_model
@pytest.fixture(
# Function scope decouples tests & allows
# env var adjustment via monkeypatch