{"record":{"id":"0c2387a2b060042c","repo":"vllm-project/vllm","slug":"encoder-cudagraph-max-vision-items-per-batch-must","errorCode":null,"errorMessage":"encoder_cudagraph_max_vision_items_per_batch must be non-negative (0 = auto-infer)","messagePattern":"encoder_cudagraph_max_vision_items_per_batch must be non-negative \\(0 = auto-infer\\)","errorType":"validation","errorClass":"ValueError","httpStatus":null,"severity":"error","filePath":"vllm/config/compilation.py","lineNumber":1038,"sourceCode":"\n        # Currently only eager and inductor backend are supported.\n        # for piecewise compilation. Custom backends are not supported for\n        # piecewise compilation. Update when more backends are supported.\n        if self.mode == CompilationMode.VLLM_COMPILE and self.backend not in [\n            \"\",\n            \"eager\",\n            \"inductor\",\n        ]:\n            raise ValueError(\n                f\"Invalid backend for piecewise compilation: {self.backend}\"\n            )\n\n        # Validate encoder CUDA graph configuration\n        if (\n            self.cudagraph_mm_encoder\n            and self.encoder_cudagraph_max_vision_items_per_batch < 0\n        ):\n            raise ValueError(\n                \"encoder_cudagraph_max_vision_items_per_batch must be \"\n                \"non-negative (0 = auto-infer)\"\n            )\n        if (\n            self.cudagraph_mm_encoder\n            and self.encoder_cudagraph_max_frames_per_batch is not None\n            and self.encoder_cudagraph_max_frames_per_batch < 0\n        ):\n            raise ValueError(\n                \"encoder_cudagraph_max_frames_per_batch must be \"\n                \"non-negative (None = auto-infer)\"\n            )\n\n        if self.encoder_cudagraph_token_budgets and any(\n            b <= 0 for b in self.encoder_cudagraph_token_budgets\n        ):\n            raise ValueError(\n                f\"All encoder_cudagraph_token_budgets must be positive, \"","sourceCodeStart":1020,"sourceCodeEnd":1056,"githubUrl":"https://github.com/vllm-project/vllm/blob/c794754062d49a8fdb63ab3c5215b488b865030c/vllm/config/compilation.py#L1020-L1056","documentation":"Error \"encoder_cudagraph_max_vision_items_per_batch must be non-negative (0 = auto-infer)\" thrown in vllm-project/vllm.","triggerScenarios":"Raised at vllm/config/compilation.py:1038 when validation fails: encoder_cudagraph_max_vision_items_per_batch must be non-negative (0 = auto-infer). 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/compilation.py:1038 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 encoder_cudagraph_max_vision_items_per_batch to a non-negative integer (0 to auto-infer)."],"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"}