vllm-project/vllm · error · ValueError

Argument {k} not found in the forward method of {cls}

Error message

Argument {k} not found in the forward method of {cls}

What it means

After dynamic dimensions are resolved (inferred or user-supplied), the decorator validates that every key of dynamic_arg_dims is an actual parameter of the decorated class's forward method. A key that does not appear in inspect.signature(cls.forward).parameters is a configuration mistake and raises ValueError naming the bad argument.

Source

Thrown at vllm/compilation/decorators.py:235

                    IntermediateTensors | None,
                ]:
                    inferred_dynamic_arg_dims[k] = 0

            logger.debug(
                ("Inferred dynamic dimensions for forward method of %s: %s"),
                cls,
                list(inferred_dynamic_arg_dims.keys()),
            )

        if len(inferred_dynamic_arg_dims) == 0:
            raise ValueError(
                "No dynamic dimensions found in the forward method of "
                f"{cls}. Please provide dynamic_arg_dims explicitly."
            )

        for k in inferred_dynamic_arg_dims:
            if k not in sig.parameters:
                raise ValueError(
                    f"Argument {k} not found in the forward method of {cls}"
                )

        return _support_torch_compile(
            cls,
            inferred_dynamic_arg_dims,
            mark_unbacked_dims,
            enable_if,
            is_encoder,
        )

    if cls is not None:
        # use `support_torch_compile` as a decorator without arguments
        assert isinstance(cls, type)
        return cls_decorator_helper(cls)

    return cls_decorator_helper

View on GitHub (pinned to c794754062)

Solutions

1. Align the dynamic_arg_dims keys with the exact parameter names in forward.
2. If the argument is forwarded via **kwargs, hoist it into an explicit named parameter.
3. Re-check after refactors that rename forward arguments.

Example fix

# before
@support_torch_compile(dynamic_arg_dims={"input_ids": 0})
class L(nn.Module):
    def forward(self, hidden_states): ...
# after
@support_torch_compile(dynamic_arg_dims={"hidden_states": 0})
class L(nn.Module):
    def forward(self, hidden_states): ...

import inspect

def validate_dynamic_arg_dims(cls, dims: dict) -> list[str]:
    params = set(inspect.signature(cls.forward).parameters)
    return [k for k in dims if k not in params]  # must be empty
assert not validate_dynamic_arg_dims(MyLayer, {'hidden_states': 0})

When it happens

Trigger: Passing dynamic_arg_dims={'input_ids': 0} to @support_torch_compile when forward has no parameter named input_ids (e.g. it is called hidden_states or embedded in **kwargs).

Common situations: Renaming forward parameters without updating dynamic_arg_dims; copying a decorator config from another model with a different signature; parameter swallowed by **kwargs so it never appears in the signature.

Related errors

• No dynamic dimensions found in the forward method of {cls}.
• decorated class should have a forward method.
• {init} received a positional argument of type {arg_type}, bu
• shape_id='{shape_id}' requires PyTorch >= 2.11.0
• Unsupported dynamic dimensions {dims} for argument {k} with

AI-assisted analysis of vllm-project/vllm@c794754062 (2026-08-14). Data as JSON: /api/errors/8511885505257aaa. Report an issue: GitHub.