vllm-project/vllm · error · ValueError

CUDAGraphMode.{cudagraph_mode.name} is not supported with {m

Error message

CUDAGraphMode.{cudagraph_mode.name} is not supported with {min_cg_attn_backend} backend (support:{min_cg_support}) ; please try cudagraph_mode=PIECEWISE, and make sure compilation mode is VLLM_COMPILE

What it means

Error "CUDAGraphMode.{cudagraph_mode.name} is not supported with {min_cg_attn_backend} backend (support:{min_cg_support}) ; please try cudagraph_mode=PIECEWISE, and make sure compilation mode is VLLM_COMPILE" thrown in vllm-project/vllm.

Source

Thrown at vllm/config/compilation.py:1467

                f"CUDAGraphMode.{cudagraph_mode.name} is not supported"
                f" with spec-decode for attention backend "
                f"{min_cg_attn_backend} (support: {min_cg_support})"
            )
            if self.splitting_ops_contain_attention():
                msg += "; setting cudagraph_mode=PIECEWISE"
                cudagraph_mode = CUDAGraphMode.PIECEWISE
            else:
                msg += "; setting cudagraph_mode=NONE"
                cudagraph_mode = CUDAGraphMode.NONE
            logger.warning(msg)

        # double check that we can support full cudagraph if they are requested
        # even after automatic downgrades
        if (
            cudagraph_mode.has_full_cudagraphs()
            and min_cg_support == AttentionCGSupport.NEVER
        ):
            raise ValueError(
                f"CUDAGraphMode.{cudagraph_mode.name} is not "
                f"supported with {min_cg_attn_backend} backend ("
                f"support:{min_cg_support}) "
                "; please try cudagraph_mode=PIECEWISE, "
                "and make sure compilation mode is VLLM_COMPILE"
            )

        # MRV1 adjusts cudagraph sizes to be a multiple of uniform_decode_query_len
        # to avoid: https://github.com/vllm-project/vllm/issues/28207 and temp-fix:
        # https://github.com/vllm-project/vllm/issues/28207#issuecomment-3504004536
        # Will be removed in the near future when we have separate cudagraph capture
        # sizes for decode and mixed prefill-decode.
        # MRV2 handles cudagraph capture sizing in cudagraph_utils.py
        # and doesn't need below: https://github.com/vllm-project/vllm/pull/45953
        if (
            not use_v2_model_runner
            and cudagraph_mode.decode_mode() == CUDAGraphMode.FULL
            and uniform_decode_query_len > 1

View on GitHub (pinned to c794754062)

Solutions

1. Set cudagraph_mode=PIECEWISE in the compilation config.
2. Ensure compilation mode is VLLM_COMPILE (-O3).
3. Or select an attention backend with stronger CUDA graph support.

When it happens

Trigger: Raised at vllm/config/compilation.py:1467 when validation fails: CUDAGraphMode not supported with the selected attention backend; try cudagraph_mode=PIECEWISE with compilation mode VLLM_COMPILE. 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/config/compilation.py:1467 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/068fcf03e462d1f1. Report an issue: GitHub.