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

55
tests/quantization/test_blackwell_moe.py
View File

@@ -11,8 +11,9 @@ from tests.utils import RemoteOpenAIServer
from vllm.platforms import current_platform
if not current_platform.is_device_capability(100):
pytest.skip("This test only runs on Blackwell GPUs (SM100).",
allow_module_level=True)
pytest.skip(
"This test only runs on Blackwell GPUs (SM100).", allow_module_level=True
)
os.environ["FLASHINFER_NVCC_THREADS"] = "16"
@@ -22,7 +23,6 @@ dummy_hf_overrides = {"num_layers": 4, "num_hidden_layers": 4}
def can_initialize(model: str, extra_args: Optional[list[str]] = None):
# Server arguments
extra_args = extra_args if extra_args is not None else []
server_args = [
@@ -40,10 +40,11 @@ def can_initialize(model: str, extra_args: Optional[list[str]] = None):
# Launch server and make a simple request
with RemoteOpenAIServer(
model,
server_args,
max_wait_seconds=1000, # Due to FlashInfer compile
override_hf_configs=dummy_hf_overrides) as server:
model,
server_args,
max_wait_seconds=1000, # Due to FlashInfer compile
override_hf_configs=dummy_hf_overrides,
) as server:
client = server.get_client()
# Make a simple request to verify the server works
completion = client.completions.create(
@@ -59,20 +60,21 @@ def can_initialize(model: str, extra_args: Optional[list[str]] = None):
## Llama4 ##
@pytest.mark.skip(reason=(
"RuntimeError: run_moe() Expected a value of type "
"'Optional[List[Tensor]]' for argument '_9' but instead found type "
"'list'."))
def test_llama4_fp8_tensor_moe_flashinfer_cutlass(
monkeypatch: pytest.MonkeyPatch):
@pytest.mark.skip(
reason=(
"RuntimeError: run_moe() Expected a value of type "
"'Optional[List[Tensor]]' for argument '_9' but instead found type "
"'list'."
)
)
def test_llama4_fp8_tensor_moe_flashinfer_cutlass(monkeypatch: pytest.MonkeyPatch):
monkeypatch.setenv("VLLM_USE_FLASHINFER_MOE_FP8", "1")
monkeypatch.setenv("VLLM_FLASHINFER_MOE_BACKEND", "throughput")
can_initialize("nvidia/Llama-4-Scout-17B-16E-Instruct-FP8")
@pytest.mark.skip(reason="Works, but takes too long to run")
def test_llama4_fp8_tensor_moe_flashinfer_trtllm(
monkeypatch: pytest.MonkeyPatch):
def test_llama4_fp8_tensor_moe_flashinfer_trtllm(monkeypatch: pytest.MonkeyPatch):
monkeypatch.setenv("VLLM_USE_FLASHINFER_MOE_FP8", "1")
monkeypatch.setenv("VLLM_FLASHINFER_MOE_BACKEND", "latency")
can_initialize("nvidia/Llama-4-Scout-17B-16E-Instruct-FP8")
@@ -100,24 +102,25 @@ def test_deepseek_fp8_block_moe_deep_gemm(monkeypatch: pytest.MonkeyPatch):
can_initialize("deepseek-ai/DeepSeek-V3.1")
@pytest.mark.skip(reason=("Known issue: lack of kernel support. "
"Expected failure: assert self.block_quant is None"))
def test_deepseek_fp8_block_moe_flashinfer_cutlass(
monkeypatch: pytest.MonkeyPatch):
@pytest.mark.skip(
reason=(
"Known issue: lack of kernel support. "
"Expected failure: assert self.block_quant is None"
)
)
def test_deepseek_fp8_block_moe_flashinfer_cutlass(monkeypatch: pytest.MonkeyPatch):
monkeypatch.setenv("VLLM_USE_FLASHINFER_MOE_FP8", "1")
monkeypatch.setenv("VLLM_FLASHINFER_MOE_BACKEND", "throughput")
can_initialize("deepseek-ai/DeepSeek-V3.1")
def test_deepseek_fp8_block_moe_flashinfer_trtllm(
monkeypatch: pytest.MonkeyPatch):
def test_deepseek_fp8_block_moe_flashinfer_trtllm(monkeypatch: pytest.MonkeyPatch):
monkeypatch.setenv("VLLM_USE_FLASHINFER_MOE_FP8", "1")
monkeypatch.setenv("VLLM_FLASHINFER_MOE_BACKEND", "latency")
can_initialize("deepseek-ai/DeepSeek-V3.1")
def test_deepseek_nvfp4_moe_flashinfer_cutlass(
monkeypatch: pytest.MonkeyPatch):
def test_deepseek_nvfp4_moe_flashinfer_cutlass(monkeypatch: pytest.MonkeyPatch):
monkeypatch.setenv("VLLM_USE_FLASHINFER_MOE_FP4", "1")
monkeypatch.setenv("VLLM_FLASHINFER_MOE_BACKEND", "throughput")
can_initialize("nvidia/DeepSeek-R1-0528-FP4-v2")
@@ -138,13 +141,11 @@ def test_gptoss_mxfp4bf16_moe_flashinfer(monkeypatch: pytest.MonkeyPatch):
can_initialize("openai/gpt-oss-20b")
def test_gptoss_mxfp4mxfp8_moe_flashinfer_cutlass(
monkeypatch: pytest.MonkeyPatch):
def test_gptoss_mxfp4mxfp8_moe_flashinfer_cutlass(monkeypatch: pytest.MonkeyPatch):
monkeypatch.setenv("VLLM_USE_FLASHINFER_MOE_MXFP4_MXFP8_CUTLASS", "1")
can_initialize("openai/gpt-oss-20b")
def test_gptoss_mxfp4mxfp8_moe_flashinfer_trtllm(
monkeypatch: pytest.MonkeyPatch):
def test_gptoss_mxfp4mxfp8_moe_flashinfer_trtllm(monkeypatch: pytest.MonkeyPatch):
monkeypatch.setenv("VLLM_USE_FLASHINFER_MOE_MXFP4_MXFP8", "1")
can_initialize("openai/gpt-oss-20b")