sgl-project/sglang · error · ValueError

MPS currently supports only --num-gpus 1

Error message

MPS currently supports only --num-gpus 1

What it means

On Apple MPS there is no multi-GPU support, so the runtime enforces --num-gpus 1 during platform-specific adjustment.

Source

Thrown at python/sglang/multimodal_gen/runtime/server_args/server_args.py:1422

    @staticmethod
    def _is_ltx23_model_path(model_path: str | None) -> bool:
        if not model_path:
            return False
        normalized = model_path.lower()
        return any(
            token in normalized
            for token in (
                "lightricks/ltx-2.3",
                "models--lightricks--ltx-2.3",
                "lightricks__ltx-2.3",
            )
        )

    def _adjust_platform_specific(self):
        if current_platform.is_mps():
            if self.num_gpus != 1:
                raise ValueError("MPS currently supports only --num-gpus 1")
            if self.component_residency is not None and any(
                mode not in (RESIDENT, LAYERWISE_OFFLOAD)
                for mode in self.component_residency.values()
            ):
                raise ValueError(
                    "MPS supports only resident or layerwise-offload component "
                    "residency"
                )
            self.use_fsdp_inference = False

    def is_arg_explicitly_set(self, arg_name: str) -> bool:
        return arg_name in self._explicit_arg_names

    def canonical_residency_mode(self, component_name: str) -> str | None:
        """Resolve the canonical selector for one component, if present."""
        return resolve_component_residency_mode(
            component_name, self.component_residency
        )

View on GitHub (pinned to 0132848349)

Solutions

1. Set --num-gpus 1 explicitly on MPS
2. Use a platform-conditional launch config
3. Move multi-GPU workloads to a CUDA host

Example fix

# before
--num-gpus 2  # on Apple Silicon
# after
--num-gpus 1

if platform.system() == 'Darwin':
    assert num_gpus == 1, 'MPS supports only 1 GPU'

When it happens

Trigger: Running on macOS with current_platform.is_mps() true while --num-gpus > 1 (or a config defaulting num_gpus to the machine's GPU-count heuristic).

Common situations: Reusing a Linux multi-GPU config on a Mac; a launcher auto-setting num_gpus = torch.cuda.device_count()-like value that is >1 on MPS.

Related errors

• MPS supports only resident or layerwise-offload component re
• This browser cannot encode H.264 MP4
• H.264 encoder did not return MP4 decoder config
• This browser does not support gzip stream decoding
• delta payload size mismatch: expected ${expectedSize}, got $

AI-assisted analysis of sgl-project/sglang@0132848349 (2026-08-28). Data as JSON: /api/errors/8de650311176b95b. Report an issue: GitHub.