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

[Distributed] Basic set of configuration for large EP deployment on GB200 (#27328)

Browse Source 
Signed-off-by: Pengchao Wang <wpc@fb.com>
Co-authored-by: Lu Fang <30275821+houseroad@users.noreply.github.com>
This commit is contained in:
Pengchao Wang
2025-10-24 14:16:44 -07:00
committed by GitHub
parent 0402428200
commit d95d0f4b98
2 changed files with 25 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
vllm/distributed/device_communicators/all2all.py
View File

@@ -277,7 +277,7 @@ class DeepEPHTAll2AllManager(DeepEPAll2AllManagerBase):
num_rdma_bytes = None
num_qps_per_rank = None
if self.internode:
if self.internode and not envs.VLLM_DEEPEP_HIGH_THROUGHPUT_FORCE_INTRA_NODE:
num_rdma_bytes = envs.VLLM_DEEPEP_BUFFER_SIZE_MB * 1024 * 1024
num_qps_per_rank = self.num_sms // 2
else:
@@ -363,6 +363,8 @@ 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=envs.VLLM_DEEPEP_LOW_LATENCY_ALLOW_NVLINK,
allow_mnnvl=envs.VLLM_DEEPEP_LOW_LATENCY_USE_MNNVL,
)
def get_handle(self, kwargs):