jax-ml/jax · error · TypeError

Expected kind to be a dtype, string, or tuple; got {kind=}

Error message

Expected kind to be a dtype, string, or tuple; got {kind=}

What it means

jax.numpy.isdtype(dtype, kind) requires `kind` to be a dtype, a string category (like 'real','numeric','signed integer'), or a (possibly nested) tuple of those. This TypeError is raised when `kind` is some other object (e.g. a list, an int, a class instance) and NumPy's np.dtype(kind) also raised TypeError. It mirrors the semantics of the NumPy NEP 49 isdtype extension.

Source

Thrown at jax/_src/dtypes.py:675

    True or False
  """
  the_dtype = np.dtype(dtype)
  kind_tuple: tuple[str | DTypeLike, ...] = (
    kind if isinstance(kind, tuple) else (kind,)
  )
  options: set[DType] = set()
  for kind in kind_tuple:
    if isinstance(kind, str) and kind in _dtype_kinds:
      options.update(_dtype_kinds[kind])
      continue
    try:
      _dtype = np.dtype(kind)
    except TypeError as e:
      if isinstance(kind, str):
        raise ValueError(
          f"Unrecognized {kind=} expected one of {list(_dtype_kinds.keys())}, "
          "or a compatible input for jnp.dtype()")
      raise TypeError(
        f"Expected kind to be a dtype, string, or tuple; got {kind=}"
      ) from e
    options.add(_dtype)
  return the_dtype in options


def _jax_type(dtype: DType, weak_type: bool) -> JAXType:
  """Return the jax type for a dtype and weak type."""
  if weak_type:
    if dtype == bool:
      return dtype
    if dtype in _custom_float_dtypes:
      return float
    return type(dtype.type(0).item())
  return dtype

def _dtype_and_weaktype(value: Any) -> tuple[DType, bool]:
  """Return a (dtype, weak_type) tuple for the given input."""

View on GitHub (pinned to 1e1c6a8fc0)

Solutions

1. Convert the kind argument to a tuple: jnp.isdtype(dt, tuple(kinds)) if it is a list
2. Use one of the recognized kind strings: 'bool','signed integer','unsigned integer','integral','real floating','complex floating','numeric','real'
3. Pass a concrete dtype (e.g. np.dtype('float32') or jnp.float32) as kind

Example fix

# before
jnp.isdtype(x.dtype, kind=['real', 'complex'])

# after
jnp.isdtype(x.dtype, kind=('real', 'complex'))

def valid_kind(k):
    from jax._src import dtypes
    allowed = {'bool','signed integer','unsigned integer','integral',
               'real floating','complex floating','numeric','real'}
    if isinstance(k, tuple):
        return all(valid_kind(x) for x in k)
    return isinstance(k, str) and k in allowed or hasattr(k, 'itemsize') or k in dtypes._dtype_kinds
kind = tuple(kind) if isinstance(kind, list) else kind

def is_isdtype_kind(k: object) -> bool:
    if isinstance(k, tuple):
        return all(is_isdtype_kind(x) for x in k)
    if isinstance(k, str):
        return k in {'bool','signed integer','unsigned integer','integral',
                     'real floating','complex floating','numeric','real'}
    return isinstance(k, np.dtype) or k in (bool, int, float, complex)

When it happens

Trigger: Calling jnp.isdtype(x.dtype, kind=["real","complex"]) (list instead of tuple), passing a dtype class object like kind=int32-class where np.dtype fails, or passing an arbitrary Python object as kind. Strings that fail np.dtype parsing raise ValueError (the sibling branch); this TypeError is specifically for non-str non-dtype objects.

Common situations: Porting NumPy code where a list was used for kind; building kind dynamically from user input that arrives as a list; passing a numpy scalar instance or a shape tuple by mistake.

Understand the failure class

Background: Invalid argument type errors: "must be of type string", "expected X, got Y", and ERR_INVALID_ARG_TYPE explained — this error's family across 15 libraries.

Related errors

• dtype must be specified.
• out_dtype should be an integer type; got {out_dtype}
• out_dtype must be integer typed; got {out_dtype=}
• primal and tangent arguments to jax.jvp do not match; dtypes
• unexpected JAX type (e.g. shape/dtype) for gradient ref pass

AI-assisted analysis of jax-ml/jax@1e1c6a8fc0 (2026-08-27). Data as JSON: /api/errors/46829c1514a494a2. Report an issue: GitHub.