[Test] Refactor hard coded device string in test files under compile/quantization/models/model_executor folders (#38901)

Signed-off-by: Liao, Wei <wei.liao@intel.com>
This commit is contained in:
wliao2
2026-04-15 11:02:35 +08:00
committed by GitHub
parent f4b42df048
commit 3abf858443
24 changed files with 122 additions and 66 deletions
@@ -35,6 +35,8 @@ from vllm.platforms import current_platform
from vllm.utils.system_utils import update_environment_variables
from vllm.utils.torch_utils import set_random_seed
DEVICE_TYPE = current_platform.device_type
pytestmark = pytest.mark.skipif(not current_platform.is_cuda(), reason="Only test CUDA")
FP8_DTYPE = current_platform.fp8_dtype()
@@ -228,7 +230,7 @@ def sequence_parallelism_pass_on_test_model(
):
set_random_seed(0)
device = torch.device(f"cuda:{local_rank}")
device = torch.device(f"{DEVICE_TYPE}:{local_rank}")
torch.accelerator.set_device_index(device)
torch.set_default_device(device)
torch.set_default_dtype(dtype)
@@ -258,7 +260,7 @@ def sequence_parallelism_pass_on_test_model(
eliminate_noops=True,
),
) # NoOp needed for fusion
device_config = DeviceConfig(device=torch.device("cuda"))
device_config = DeviceConfig(device=torch.device(DEVICE_TYPE))
# this is a fake model name to construct the model config
# in the vllm_config, it's not really used.