vllm-project/vllm · error · ValueError

Stochastic rounding for Mamba cache with triton backend requ

Error message

Stochastic rounding for Mamba cache with triton backend requires compute capability 10.0 (data center Blackwell). The `cvt.rs` PTX instruction is not supported on your GPU. Please do not specify `--enable-mamba-cache-stochastic-rounding`, or set `--mamba-backend flashinfer`.

What it means

Error "Stochastic rounding for Mamba cache with triton backend requires compute capability 10.0 (data center Blackwell). The `cvt.rs` PTX instruction is not supported on your GPU. Please do not specify `--enable-mamba-cache-stochastic-rounding`, or set `--mamba-backend flashinfer`." thrown in vllm-project/vllm.

Source

Thrown at vllm/config/mamba.py:98

                "or omit `--mamba-ssu-algorithm`."
            )

    def __post_init__(self):
        self.validate_ssu_algorithm()
        if self.enable_stochastic_rounding:
            from vllm.platforms import current_platform

            if not current_platform.is_cuda():
                raise ValueError(
                    "Stochastic rounding for Mamba cache is only supported "
                    "on NVIDIA CUDA platforms. Please do not specify  "
                    "`--enable-mamba-cache-stochastic-rounding`."
                )
            if (
                self.backend == MambaBackendEnum.TRITON
                and not current_platform.is_device_capability_family(100)
            ):
                raise ValueError(
                    "Stochastic rounding for Mamba cache with triton backend requires "
                    "compute capability 10.0 (data center Blackwell). The `cvt.rs` "
                    "PTX instruction is not supported on your GPU. Please do not "
                    "specify `--enable-mamba-cache-stochastic-rounding`, "
                    "or set `--mamba-backend flashinfer`."
                )

View on GitHub (pinned to c794754062)

Solutions

1. Do not specify --enable-mamba-cache-stochastic-rounding on this GPU.
2. Or set --mamba-backend flashinfer.
3. Or run on a compute capability 10.0 (Blackwell) GPU.

When it happens

Trigger: Raised at vllm/config/mamba.py:98 when validation fails: Stochastic rounding for Mamba cache with triton backend requires compute capability 10.0. 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/mamba.py:98 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/986ebbbf1dc2daf6. Report an issue: GitHub.