vllm-project/vllm · error · RuntimeError

EPLB communicator 'pynccl' requested but expert weights cont

Error message

EPLB communicator 'pynccl' requested but expert weights contain unsupported dtypes: ({', '.join(str(dtype) for dtype in unsupported_dtypes)}).

What it means

Error "EPLB communicator 'pynccl' requested but expert weights contain unsupported dtypes: ({', '.join(str(dtype) for dtype in unsupported_dtypes)})." thrown in vllm-project/vllm.

Source

Thrown at vllm/distributed/eplb/eplb_communicator.py:708

        else group_coordinator.device_group
    )

    def _create_pynccl() -> EplbCommunicator:
        if tensor_device_type == "cpu":
            raise RuntimeError(
                "EPLB communicator 'pynccl' supports only cuda-like devices "
                f"(got {tensor_device_type})."
            )
        unsupported_dtypes = sorted(
            {
                tensor.dtype
                for tensor in first_layer
                if not ncclDataTypeEnum.supports_torch_dtype(tensor.dtype)
            },
            key=str,
        )
        if unsupported_dtypes:
            raise RuntimeError(
                "EPLB communicator 'pynccl' requested but expert weights contain "
                "unsupported dtypes: "
                f"({', '.join(str(dtype) for dtype in unsupported_dtypes)})."
            )

        device_comm = group_coordinator.device_communicator
        pynccl_comm = (
            getattr(device_comm, "pynccl_comm", None)
            if device_comm is not None
            else None
        )
        if pynccl_comm is None or pynccl_comm.disabled or not pynccl_comm.available:
            raise RuntimeError("EPLB communicator 'pynccl' requested but unavailable.")
        try:
            return PyNcclEplbCommunicator(pynccl_comm=pynccl_comm)
        except Exception as exc:
            raise RuntimeError(
                f"Failed to initialize PyNcclEplbCommunicator ({exc})."

View on GitHub (pinned to c794754062)

Solutions

1. Convert expert weights to dtypes supported by pynccl, or choose a different EPLB communicator backend.

When it happens

Trigger: Raised at vllm/distributed/eplb/eplb_communicator.py:708 when validation fails: EPLB communicator 'pynccl' requested but expert weights contain unsupported dtypes. 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:708 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/83e58f49d8b8dc34. Report an issue: GitHub.