[Bugfix] Pin EPLB and MLA host-to-device transfer buffers (#56138)

Signed-off-by: khluu <khluu000@gmail.com>
Signed-off-by: Kevin H. Luu <khluu000@gmail.com>
Co-authored-by: OpenAI Codex <noreply@openai.com>
This commit is contained in:
Kevin H. Luu
2026-09-11 01:57:47 -07:00
committed by GitHub
co-authored by OpenAI Codex
parent d0dfe587d5
commit 1cf6555214
4 changed files with 81 additions and 5 deletions
+4 -1
View File
@@ -34,6 +34,7 @@ from vllm.distributed.utils import is_weak_contiguous
from vllm.logger import init_logger
from vllm.platforms import current_platform
from vllm.utils.gpu_sync_debug import gpu_sync_allowed
from vllm.utils.torch_utils import PIN_MEMORY
logger = init_logger(__name__)
@@ -204,7 +205,9 @@ class TorchDistGlooStagedEplbCommunicator(EplbCommunicator):
)
)
continue
cpu_tensor = torch.empty_like(tensor, device="cpu")
cpu_tensor = torch.empty_like(
tensor, device="cpu", pin_memory=PIN_MEMORY
)
p2p_ops.append(
P2POp(
torch.distributed.irecv,