[ROCm] ROCm DeepEP API updated to latest (#39721)

Signed-off-by: Tej Kiran <vpolamre@amd.com>
Signed-off-by: tej <37236721+itej89@users.noreply.github.com>
Co-authored-by: TJian <tunjian.tan@embeddedllm.com>
Co-authored-by: HAIAI <39548240+HAIAI@users.noreply.github.com>
This commit is contained in:
tej
2026-04-30 07:46:59 -07:00
committed by GitHub
co-authored by TJian HAIAI
parent f03d82efdd
commit 121dbe7a22
3 changed files with 42 additions and 60 deletions
@@ -10,7 +10,6 @@ import vllm.envs as envs
from vllm.distributed import get_dp_group, get_ep_group
from vllm.forward_context import get_forward_context
from vllm.logger import init_logger
from vllm.platforms import current_platform
from vllm.utils.flashinfer import (
has_flashinfer_nvlink_one_sided,
has_flashinfer_nvlink_two_sided,
@@ -225,11 +224,8 @@ class DeepEPHTAll2AllManager(DeepEPAll2AllManagerBase):
num_rdma_bytes=num_rdma_bytes,
low_latency_mode=False,
num_qps_per_rank=num_qps_per_rank,
explicitly_destroy=True,
)
if not current_platform.is_rocm():
kwargs.update(
explicitly_destroy=True,
)
return kwargs
def get_handle(self, kwargs):
@@ -303,13 +299,10 @@ class DeepEPLLAll2AllManager(DeepEPAll2AllManagerBase):
num_rdma_bytes=num_rdma_bytes,
low_latency_mode=True,
num_qps_per_rank=num_qps_per_rank,
allow_nvlink_for_low_latency_mode=True,
allow_mnnvl=envs.VLLM_DEEPEP_LOW_LATENCY_USE_MNNVL,
explicitly_destroy=True,
)
if not current_platform.is_rocm():
kwargs.update(
allow_nvlink_for_low_latency_mode=True,
allow_mnnvl=envs.VLLM_DEEPEP_LOW_LATENCY_USE_MNNVL,
explicitly_destroy=True,
)
return kwargs
def get_handle(self, kwargs):