[Hardware] Replace torch.cuda.device_count/current_device/set_device API (#36145)

Signed-off-by: Kunshang Ji <jikunshang95@gmail.com>
Signed-off-by: Kunshang Ji <kunshang.ji@intel.com>
This commit is contained in:
Kunshang Ji
2026-03-12 07:57:47 -07:00
committed by GitHub
parent 2e693f48e7
commit 53ec16a705
89 changed files with 254 additions and 219 deletions
+3 -1
View File
@@ -26,7 +26,9 @@ DTYPES = [torch.half, torch.bfloat16, torch.float]
NUM_TOKENS = [7, 83, 2048] # Arbitrary values for testing
D = [512, 13824] # Arbitrary values for testing
SEEDS = [0]
CUDA_DEVICES = [f"cuda:{i}" for i in range(1 if torch.cuda.device_count() == 1 else 2)]
CUDA_DEVICES = [
f"cuda:{i}" for i in range(1 if torch.accelerator.device_count() == 1 else 2)
]
@pytest.mark.parametrize(
@@ -33,7 +33,9 @@ SCALE_UBS = [True, False]
GROUP_SIZES = [None, [1, 64], [1, 128]]
TMA_ALIGNMENTS = [0, 4]
SEEDS = [0]
CUDA_DEVICES = [f"cuda:{i}" for i in range(1 if torch.cuda.device_count() == 1 else 2)]
CUDA_DEVICES = [
f"cuda:{i}" for i in range(1 if torch.accelerator.device_count() == 1 else 2)
]
EPS = 1e-6
@@ -182,7 +184,7 @@ def test_rms_norm(
if torch.cuda.is_available():
torch.cuda.manual_seed(seed)
torch.set_default_device(device)
torch.cuda.set_device(device)
torch.accelerator.set_device_index(device)
if group_size is not None and hidden_size % group_size[1] != 0:
# skip
+3 -1
View File
@@ -14,7 +14,9 @@ NUM_TOKENS = [7, 83, 4096] # Arbitrary values for testing
HIDDEN_SIZES = [8, 768, 769, 5120, 5125, 8192] # Arbitrary values for testing
ADD_RESIDUAL = [False, True]
SEEDS = [0]
CUDA_DEVICES = [f"cuda:{i}" for i in range(1 if torch.cuda.device_count() == 1 else 2)]
CUDA_DEVICES = [
f"cuda:{i}" for i in range(1 if torch.accelerator.device_count() == 1 else 2)
]
@pytest.mark.parametrize("num_tokens", NUM_TOKENS)
+3 -1
View File
@@ -19,7 +19,9 @@ NUM_HEADS = [17] # Arbitrary values for testing
BATCH_SIZES = [5] # Arbitrary values for testing
SEQ_LENS = [11, 8192] # Arbitrary values for testing
SEEDS = [0]
CUDA_DEVICES = [f"cuda:{i}" for i in range(1 if torch.cuda.device_count() == 1 else 2)]
CUDA_DEVICES = [
f"cuda:{i}" for i in range(1 if torch.accelerator.device_count() == 1 else 2)
]
USE_KEY = [True, False]
@@ -28,7 +28,8 @@ from vllm.utils.torch_utils import set_random_seed
@pytest.mark.parametrize("block_size", [16, 64, 256])
@pytest.mark.parametrize("seed", [0])
@pytest.mark.parametrize(
"device", [f"cuda:{i}" for i in range(1 if torch.cuda.device_count() == 1 else 2)]
"device",
[f"cuda:{i}" for i in range(1 if torch.accelerator.device_count() == 1 else 2)],
)
@torch.inference_mode()
def test_concat_and_cache_mla_rope_fused(
+3 -1
View File
@@ -6,7 +6,9 @@ import torch
from vllm.utils.platform_utils import is_uva_available
from vllm.utils.torch_utils import get_accelerator_view_from_cpu_tensor
CUDA_DEVICES = [f"cuda:{i}" for i in range(1 if torch.cuda.device_count() == 1 else 2)]
CUDA_DEVICES = [
f"cuda:{i}" for i in range(1 if torch.accelerator.device_count() == 1 else 2)
]
@pytest.mark.skipif(not is_uva_available(), reason="UVA is not available.")