biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

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

Browse Source 
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 22:57:47 +08:00
committed by GitHub
parent 2e693f48e7
commit 53ec16a705
89 changed files with 254 additions and 219 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
tests/kernels/attention/test_prefix_prefill.py
View File

@@ -21,7 +21,9 @@ NUM_HEADS = [64]
NUM_QUERIES_PER_KV = [1, 64]
HEAD_SIZES = [24, 128]
DTYPES = [torch.float16]
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)
]
SLIDING_WINDOW = [0, 16, 2048]
KV_CACHE_DTYPES = ["auto", "fp8", "fp8_e5m2"]
@@ -135,7 +137,7 @@ def test_contexted_kv_attention(
# for GPU 1 would run on both GPU0 and GPU1 and things would hang
#
# see also similar issue: https://github.com/Dao-AILab/flash-attention/issues/523
torch.cuda.set_device(device)
torch.accelerator.set_device_index(device)
MAX_SEQ_LEN = 1024
MAX_CTX_LEN = 1024
@@ -356,7 +358,7 @@ def test_contexted_kv_attention_alibi(
# for GPU 1 would run on both GPU0 and GPU1 and things would hang
#
# see also similar issue: https://github.com/Dao-AILab/flash-attention/issues/523
torch.cuda.set_device(device)
torch.accelerator.set_device_index(device)
def _get_alibi_slopes(total_num_heads: int) -> torch.Tensor:
# Fork from: vllm/vllm/model_executor/models/bloom.py#L44