{"record":{"id":"de0dc4782da88777","repo":"vllm-project/vllm","slug":"inkling-lora-requires-inkling-multimem-ar-0-the-l","errorCode":null,"errorMessage":"Inkling LoRA requires INKLING_MULTIMEM_AR=0: the Lamport fused-collective path bypasses the LoRA-wrapped wo_ud and dense down_proj layers on decode-sized batches, silently dropping their LoRA. Set INKLING_MULTIMEM_AR=0.","messagePattern":"Inkling LoRA requires INKLING_MULTIMEM_AR=0: the Lamport fused-collective path bypasses the LoRA-wrapped wo_ud and dense down_proj layers on decode-sized batches, silently dropping their LoRA\\. Set INKLING_MULTIMEM_AR=0\\.","errorType":"validation","errorClass":"ValueError","httpStatus":null,"severity":"error","filePath":"vllm/config/lora.py","lineNumber":144,"sourceCode":"            raise ValueError(\"Dual CUDA streams are only supported on CUDA platforms.\")\n        if envs.VLLM_LORA_ENABLE_DUAL_STREAM and self.fully_sharded_loras:\n            logger.warning_once(\n                \"fully_sharded_loras isn't compatible with \"\n                \"VLLM_LORA_ENABLE_DUAL_STREAM, set VLLM_LORA_ENABLE_DUAL_STREAM=False\"\n            )\n            envs.VLLM_LORA_ENABLE_DUAL_STREAM = False\n        return self\n\n    def verify_with_model_config(self, model_config: ModelConfig):\n        if self.lora_dtype in (None, \"auto\"):\n            self.lora_dtype = model_config.dtype\n        elif isinstance(self.lora_dtype, str):\n            self.lora_dtype = getattr(torch, self.lora_dtype)\n\n        architectures = getattr(model_config, \"architectures\", None) or []\n        is_inkling = any(\"Inkling\" in arch for arch in architectures)\n        if is_inkling and os.environ.get(\"INKLING_MULTIMEM_AR\", \"1\") != \"0\":\n            raise ValueError(\n                \"Inkling LoRA requires INKLING_MULTIMEM_AR=0: the Lamport \"\n                \"fused-collective path bypasses the LoRA-wrapped wo_ud and dense \"\n                \"down_proj layers on decode-sized batches, silently dropping \"\n                \"their LoRA. Set INKLING_MULTIMEM_AR=0.\"\n            )\n","sourceCodeStart":126,"sourceCodeEnd":150,"githubUrl":"https://github.com/vllm-project/vllm/blob/c794754062d49a8fdb63ab3c5215b488b865030c/vllm/config/lora.py#L126-L150","documentation":"Error \"Inkling LoRA requires INKLING_MULTIMEM_AR=0: the Lamport fused-collective path bypasses the LoRA-wrapped wo_ud and dense down_proj layers on decode-sized batches, silently dropping their LoRA. Set INKLING_MULTIMEM_AR=0.\" thrown in vllm-project/vllm.","triggerScenarios":"Raised at vllm/config/lora.py:144 when validation fails: Inkling LoRA requires INKLING_MULTIMEM_AR=0. Typically triggered by an incompatible or incomplete vLLM configuration, an unsupported platform/backend combination, or a runtime resource/dependency that is missing.","commonSituations":"Commonly encountered at vllm/config/lora.py:144 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.","solutions":["Set INKLING_MULTIMEM_AR=0 in the environment when using Inkling LoRA.","Or do not use Inkling LoRA with the multimem all-reduce path."],"exampleFix":null,"handlingStrategy":null,"validationCode":null,"typeGuard":null,"tryCatchPattern":null,"preventionTips":[],"tags":[],"backgroundTag":null,"analyzedSha":"c794754062d49a8fdb63ab3c5215b488b865030c","analyzedAt":"2026-08-14T21:17:39.825Z","schemaVersion":2},"datasetVersion":"2026-08-15T22:17:37.221Z"}