A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 4, 08:45 UTC
[Bugfix][Ray] Fix RayExecutorV2 actor name collision with DP > 1 (#40398)
Signed-off-by: Tomer Asida <57313761+tomeras91@users.noreply.github.com>
tomeras91 committed
May 3, 2026 at 23:00 UTC
cb03fee32b5c191b3ac5a248e47cd58a66c75591
1 file changed
+5
vllm/v1/engine/utils.py
+5
@@ -403,6 +403,11 @@ class CoreEngineActorManager:
403
range(dp_size), local_dp_ranks, placement_groups
404
):
405
dp_vllm_config = copy.deepcopy(vllm_config)
406
+ if dp_size > 1:
407
+ # Append the DP rank to instance_id so that per-engine
408
+ # identifiers (e.g. Ray actor names in RayExecutorV2) are
409
+ # unique across DP replicas.
410
+ dp_vllm_config.instance_id = f"{dp_vllm_config.instance_id}_dp{index}"
411
dp_vllm_config.parallel_config.placement_group = pg
412
local_client = index < local_engine_count
413