[MRV2] Buffer util simplifications (#56888)

Signed-off-by: Nick Hill <nickhill123@gmail.com>
This commit is contained in:
Nick Hill
2026-09-15 20:36:51 +00:00
committed by GitHub
parent b6ce714354
commit f2aad6aa70
21 changed files with 125 additions and 112 deletions
-19
View File
@@ -10,7 +10,6 @@ from vllm.logger import init_logger
from vllm.triton_utils import tl, triton
from vllm.utils.platform_utils import is_uva_available
from vllm.utils.torch_utils import (
PIN_MEMORY,
async_tensor_h2d,
get_accelerator_view_from_cpu_tensor,
)
@@ -27,24 +26,6 @@ def set_default_max_concurrency(n: int) -> None:
_DEFAULT_MAX_CONCURRENCY = max(2, n)
def async_copy_to_gpu(
x: torch.Tensor | np.ndarray,
out: torch.Tensor | None = None,
device: torch.device | None = None,
) -> torch.Tensor:
if isinstance(x, np.ndarray):
x = torch.from_numpy(x)
assert x.is_cpu
if out is None:
assert device is not None
out = torch.empty_like(x, device=device)
# pin_memory() is no-op if the memory is already pinned.
pinned = x.pin_memory() if PIN_MEMORY else x
return out.copy_(pinned, non_blocking=True)
class UvaBuffer:
def __init__(self, size: int | Sequence[int], dtype: torch.dtype):
if not is_uva_available():