[FEAT][ROCm] Integrate Paged Attention Kernel from AITER (#15001)

Signed-off-by: vllmellm <vllm.ellm@embeddedllm.com>
Signed-off-by: tjtanaa <tunjian.tan@embeddedllm.com>
Co-authored-by: tjtanaa <tunjian.tan@embeddedllm.com>
This commit is contained in:
vllmellm
2025-04-22 02:46:28 -07:00
committed by GitHub
co-authored by tjtanaa
parent 8f7bace7c3
commit 0e237f0035
5 changed files with 195 additions and 27 deletions
+7
View File
@@ -75,6 +75,7 @@ if TYPE_CHECKING:
VLLM_DISABLED_KERNELS: list[str] = []
VLLM_USE_V1: bool = True
VLLM_ROCM_USE_AITER: bool = False
VLLM_ROCM_USE_AITER_PAGED_ATTN: bool = False
VLLM_ROCM_USE_AITER_LINEAR: bool = True
VLLM_ROCM_USE_AITER_MOE: bool = True
VLLM_ROCM_USE_AITER_RMSNORM: bool = True
@@ -533,6 +534,12 @@ environment_variables: dict[str, Callable[[], Any]] = {
lambda: (os.getenv("VLLM_ROCM_USE_AITER", "False").lower() in
("true", "1")),
# Whether to use aiter paged attention.
# By default is disabled.
"VLLM_ROCM_USE_AITER_PAGED_ATTN":
lambda: (os.getenv("VLLM_ROCM_USE_AITER_PAGED_ATTN", "False").lower() in
("true", "1")),
# use aiter linear op if aiter ops are enabled
# The following list of related ops
# - scaled_mm (per-tensor / rowwise)