From 3afe659b6bb90b961bf09984166393824a893af9 Mon Sep 17 00:00:00 2001 From: Itay Alroy <75032521+itayalroy@users.noreply.github.com> Date: Tue, 16 Jun 2026 02:37:22 +0300 Subject: [PATCH] [EP] Enable DBO with NIXL EP (#45275) Signed-off-by: Itay Alroy --- vllm/config/compilation.py | 2 +- vllm/config/vllm.py | 12 +++++++----- 2 files changed, 8 insertions(+), 6 deletions(-) diff --git a/vllm/config/compilation.py b/vllm/config/compilation.py index 6b03c7adf1..bc38ec6a8a 100644 --- a/vllm/config/compilation.py +++ b/vllm/config/compilation.py @@ -1197,7 +1197,7 @@ class CompilationConfig: "are optimized for prefill and are incompatible with CUDA Graphs. " "In order to use CUDA Graphs for decode-optimized workloads, " "use --all2all-backend with another option, such as " - "deepep_low_latency or allgather_reducescatter." + "deepep_low_latency, nixl_ep, or allgather_reducescatter." ) self.cudagraph_mode = CUDAGraphMode.NONE diff --git a/vllm/config/vllm.py b/vllm/config/vllm.py index a3bfa56f57..95e299eb02 100644 --- a/vllm/config/vllm.py +++ b/vllm/config/vllm.py @@ -1457,12 +1457,14 @@ class VllmConfig: assert a2a_backend in [ "deepep_low_latency", "deepep_high_throughput", + "nixl_ep", ], ( - "Microbatching currently only supports the deepep_low_latency and " - f"deepep_high_throughput all2all backend. {a2a_backend} is not " - "supported. To fix use --all2all-backend=deepep_low_latency or " - "--all2all-backend=deepep_high_throughput and install the DeepEP" - " kernels." + "Microbatching currently only supports the deepep_low_latency, " + "deepep_high_throughput, and nixl_ep all2all backends. " + f"{a2a_backend} is not supported. To fix use " + "--all2all-backend=deepep_low_latency, " + "--all2all-backend=deepep_high_throughput, or " + "--all2all-backend=nixl_ep and install the matching kernels." ) if not self.model_config.disable_cascade_attn: