vllm-project/vllm · error · RuntimeError

NIXL EPLB init failed: {name}

Error message

NIXL EPLB init failed: {name}

What it means

Error "NIXL EPLB init failed: {name}" thrown in vllm-project/vllm.

Source

Thrown at vllm/distributed/eplb/eplb_communicator.py:351

        """
        self._init_step("agents", self._init_remote_agents)
        self._init_step("send meta", self._exchange_remote_send_meta)
        self._remote_state_initialized = True

    def _ensure_remote_state(self) -> None:
        if not self._remote_state_initialized:
            self._init_remote_state()

    @property
    def needs_profile_buffer_reservation(self) -> bool:
        return False

    @staticmethod
    def _init_step(name: str, fn: object, *args: object, **kwargs: object) -> None:
        try:
            fn(*args, **kwargs)  # type: ignore[operator]
        except Exception as exc:
            raise RuntimeError(f"NIXL EPLB init failed: {name}") from exc

    def _make_agent_name(self) -> str:
        """Build a deployment-unique nixl agent name."""
        pp_size = get_pp_group().world_size
        pp_suffix = f"-pp{get_pp_group().rank_in_group}" if pp_size > 1 else ""
        uid = uuid.uuid4().hex[:8]
        return f"eplb-{self._rank}{pp_suffix}-{uid}"

    def set_stream(self, cuda_stream: torch.cuda.Stream | None) -> None:
        pass

    def add_send(
        self,
        tensors: list[torch.Tensor],
        dst_rank: int,
        expert_id: int,
    ) -> None:
        # No-op: NIXL READ is receiver-initiated. The sender's expert

View on GitHub (pinned to c794754062)

Solutions

1. Check NIXL installation and network configuration; the failed init step is named in the error.
2. Review NIXL logs for the underlying failure.

When it happens

Trigger: Raised at vllm/distributed/eplb/eplb_communicator.py:351 when validation fails: NIXL EPLB init failed. Typically triggered by an incompatible or incomplete vLLM configuration, an unsupported platform/backend combination, or a runtime resource/dependency that is missing.

Common situations: Commonly encountered at vllm/distributed/eplb/eplb_communicator.py:351 during vLLM startup/config validation or runtime setup when: (1) conflicting CLI flags or config fields are combined, (2) the current platform (CUDA/ROCm/CPU/XPU) or installed optional packages do not support the requested feature, or (3) a required value is absent or out of range. Resolve by correcting the configuration as described in the message, or by selecting a supported alternative.

AI-assisted analysis of vllm-project/vllm@c794754062 (2026-08-14). Data as JSON: /api/errors/f2c7216315dc746e. Report an issue: GitHub.