{"record":{"id":"e37b011d7d818118","repo":"sgl-project/sglang","slug":"weight-cache-transport-backend-vmm-fd-backend-r","errorCode":null,"errorMessage":"weight cache transport backend {VMM_FD_BACKEND!r} is not implemented in this build","messagePattern":"weight cache transport backend (.+?) is not implemented in this build","errorType":"exception","errorClass":"NotImplementedError","httpStatus":null,"severity":"error","filePath":"python/sglang/srt/weight_cache/transport.py","lineNumber":154,"sourceCode":"        return MultiprocessingSerializer.deserialize(entry[\"handle\"])\n\n\nclass VmmFdTransportBackend(WeightCacheTransportBackend):\n    \"\"\"Placeholder for the CUDA VMM + fd-passing transport.\n\n    The backend is not wired up yet: can_export_state reports False so the\n    daemon keeps selecting torch_ipc, and every other entry point fails loudly\n    instead of silently returning None.\n    \"\"\"\n\n    name = VMM_FD_BACKEND\n\n    def __init__(self):\n        self._raise_not_implemented()\n\n    @staticmethod\n    def _raise_not_implemented() -> NoReturn:\n        raise NotImplementedError(\n            f\"weight cache transport backend {VMM_FD_BACKEND!r} is not \"\n            f\"implemented in this build\"\n        )\n\n    @classmethod\n    def can_export_state(\n        cls, state_tensors: Mapping[str, Tuple[torch.Tensor, bool]]\n    ) -> bool:\n        return False\n\n    def prepare_export(\n        self, state_tensors: Mapping[str, Tuple[torch.Tensor, bool]]\n    ) -> Dict[str, Dict[str, Any]]:\n        self._raise_not_implemented()\n\n    def send_fetch_state_response(\n        self,\n        conn: socket.socket,","sourceCodeStart":136,"sourceCodeEnd":172,"githubUrl":"https://github.com/sgl-project/sglang/blob/0132848349585cfe6aae51c4941cbae872505f8a/python/sglang/srt/weight_cache/transport.py#L136-L172","documentation":"VmmFdTransportBackend (the 'vmm_fd' weight cache transport) is a stub in this build: every method raises NotImplementedError. Selected when the backend name is vmm_fd, but the feature was not compiled/implemented here.","triggerScenarios":"Setting the weight cache transport backend to 'vmm_fd' via config/env; any call to its __init__, prepare_export, send/recv_fetch_state_response, or import_tensor.","commonSituations":"Using a docs/blog config that references the experimental VMM FD backend on a release that lacks it.","solutions":["Use the default torch IPC backend (unset the backend name)","Upgrade sglang to a build that implements vmm_fd"],"exampleFix":"# before\nbackend = get_client_transport_backend(\"vmm_fd\")\n# after\nbackend = get_client_transport_backend(None)  # torch IPC","handlingStrategy":"validation","validationCode":"from sglang.srt.weight_cache.transport import TORCH_IPC_BACKEND, VMM_FD_BACKEND\nassert name in (None, TORCH_IPC_BACKEND) or implemented(name)","typeGuard":null,"tryCatchPattern":null,"preventionTips":["Don't select vmm_fd unless your build implements it","Default to torch_ipc backend"],"tags":["transport","not-implemented","weight-cache","config"],"backgroundTag":"feature-not-implemented","analyzedSha":"0132848349585cfe6aae51c4941cbae872505f8a","analyzedAt":"2026-08-28T05:10:05.995Z","schemaVersion":2},"datasetVersion":"2026-08-28T06:17:29.519Z"}