mirror of
https://git.datalinker.icu/vllm-project/vllm.git
synced 2026-05-26 00:57:53 +08:00
[CI/Build][AMD] Skip tests in test_fusions_e2e and test_dbo_dp_ep_gsm8k that require non-existing imports for ROCm (#30417)
Signed-off-by: Randall Smith <ransmith@amd.com> Co-authored-by: Randall Smith <ransmith@amd.com>
This commit is contained in:
parent
d527cf0b3d
commit
48661d275f
@ -138,6 +138,17 @@ elif current_platform.is_rocm():
|
|||||||
CUSTOM_OPS_FP8 = ["-quant_fp8", "+quant_fp8"]
|
CUSTOM_OPS_FP8 = ["-quant_fp8", "+quant_fp8"]
|
||||||
|
|
||||||
|
|
||||||
|
def has_cuda_graph_wrapper_metadata() -> bool:
|
||||||
|
from importlib import import_module
|
||||||
|
|
||||||
|
try:
|
||||||
|
module = import_module("torch._inductor.utils")
|
||||||
|
module.CUDAGraphWrapperMetadata # noqa B018
|
||||||
|
except AttributeError:
|
||||||
|
return False
|
||||||
|
return True
|
||||||
|
|
||||||
|
|
||||||
@pytest.mark.parametrize(
|
@pytest.mark.parametrize(
|
||||||
"model_name, model_kwargs, backend, matches, custom_ops",
|
"model_name, model_kwargs, backend, matches, custom_ops",
|
||||||
# Test attention+quant_fp8 fusion with custom and torch impls of QuantFP8
|
# Test attention+quant_fp8 fusion with custom and torch impls of QuantFP8
|
||||||
@ -145,7 +156,20 @@ CUSTOM_OPS_FP8 = ["-quant_fp8", "+quant_fp8"]
|
|||||||
# quant_fp4 only has the custom impl
|
# quant_fp4 only has the custom impl
|
||||||
+ list(flat_product(MODELS_FP4, [""])),
|
+ list(flat_product(MODELS_FP4, [""])),
|
||||||
)
|
)
|
||||||
@pytest.mark.parametrize("inductor_graph_partition", [True, False])
|
@pytest.mark.parametrize(
|
||||||
|
"inductor_graph_partition",
|
||||||
|
[
|
||||||
|
pytest.param(
|
||||||
|
True,
|
||||||
|
marks=pytest.mark.skipif(
|
||||||
|
not has_cuda_graph_wrapper_metadata(),
|
||||||
|
reason="This test requires"
|
||||||
|
"torch._inductor.utils.CUDAGraphWrapperMetadata to run",
|
||||||
|
),
|
||||||
|
),
|
||||||
|
False,
|
||||||
|
],
|
||||||
|
)
|
||||||
def test_attn_quant(
|
def test_attn_quant(
|
||||||
model_name: str,
|
model_name: str,
|
||||||
model_kwargs: dict[str, Any],
|
model_kwargs: dict[str, Any],
|
||||||
|
|||||||
@ -13,6 +13,7 @@ import torch
|
|||||||
|
|
||||||
from tests.evals.gsm8k.gsm8k_eval import evaluate_gsm8k
|
from tests.evals.gsm8k.gsm8k_eval import evaluate_gsm8k
|
||||||
from tests.utils import RemoteOpenAIServer
|
from tests.utils import RemoteOpenAIServer
|
||||||
|
from vllm.utils.import_utils import has_deep_ep
|
||||||
|
|
||||||
# Detect Blackwell / B200 (compute capability 10.x)
|
# Detect Blackwell / B200 (compute capability 10.x)
|
||||||
try:
|
try:
|
||||||
@ -44,6 +45,7 @@ DEEPEP_BACKENDS = [
|
|||||||
]
|
]
|
||||||
|
|
||||||
|
|
||||||
|
@pytest.mark.skipif(not has_deep_ep(), reason="These tests require deep_ep to run")
|
||||||
@pytest.mark.parametrize("all2all_backend", DEEPEP_BACKENDS)
|
@pytest.mark.parametrize("all2all_backend", DEEPEP_BACKENDS)
|
||||||
@pytest.mark.xfail(
|
@pytest.mark.xfail(
|
||||||
IS_BLACKWELL,
|
IS_BLACKWELL,
|
||||||
|
|||||||
Loading…
x
Reference in New Issue
Block a user