vllm-project/vllm · error · ValueError

capture_torch_profiler is only applicable when profiler is s

Error message

capture_torch_profiler is only applicable when profiler is set to 'torch'

What it means

ProfilerConfig rejects capture_torch_profiler=True when profiler is not 'torch'. The flag enables a torch profiler specifically during CUDA graph capture on rank 0 and writes traces into a capture_traces subdirectory of torch_profiler_dir, so it is meaningless without the torch profiler selected.

Source

Thrown at vllm/config/profiler.py:232

        if proton_profiler_dir:
            if _is_uri_path(proton_profiler_dir):
                raise ValueError("proton_profiler_dir must be a local directory")
            self.proton_profiler_dir = os.path.abspath(
                os.path.expanduser(proton_profiler_dir)
            )

        if self.profiler == "proton":
            output_format = self.proton_output_format
            if output_format == "chrome_trace" and self.proton_data != "trace":
                raise ValueError("chrome_trace output requires proton_data='trace'")
            if (
                output_format in ("hatchet", "hatchet_msgpack")
                and self.proton_data != "tree"
            ):
                raise ValueError(f"{output_format} output requires proton_data='tree'")

        if self.capture_torch_profiler and self.profiler != "torch":
            raise ValueError(
                "capture_torch_profiler is only applicable when profiler is "
                "set to 'torch'"
            )

        return self

View on GitHub (pinned to c794754062)

Solutions

1. Set profiler='torch' (and torch_profiler_dir) when using capture_torch_profiler.
2. Or remove/disable capture_torch_profiler if not profiling with torch.

Example fix

# before
ProfilerConfig(capture_torch_profiler=True)  # profiler is None

# after
ProfilerConfig(profiler='torch', torch_profiler_dir='/tmp/tr', capture_torch_profiler=True)

def validate_capture_flag(capture: bool, profiler: str | None) -> None:
    if capture and profiler != "torch":
        raise SystemExit("capture_torch_profiler requires profiler='torch' (and torch_profiler_dir)")

When it happens

Trigger: Passing --profiler-config capture_torch_profiler=true with profiler unset, or profiler='proton'/'cuda'.

Common situations: Debugging CUDA graph capture with a copied flag set from a torch-profiler config; enabling all profiling flags defensively.

Related errors

• Unable to use nsight profiling unless workers run with Ray.
• torch_profiler_dir is only applicable when profiler is set t
• torch_profiler_dir must be set when profiler is 'torch'
• {options} only applicable when profiler is set to 'proton'
• proton_profiler_dir must be set when profiler is 'proton'

AI-assisted analysis of vllm-project/vllm@c794754062 (2026-08-14). Data as JSON: /api/errors/083a663d0b1701db. Report an issue: GitHub.