jax-ml/jax · error · NotImplementedError

The 'out' argument to jnp.var is not supported.

Error message

The 'out' argument to jnp.var is not supported.

What it means

JAX arrays are immutable, so var cannot write its result into a caller-provided out buffer. The jitted _var helper fires when out is not None.

Solutions

  1. Use the returned value: v = jnp.var(x, axis=0)
  2. Update buffers with .at[...].set(jnp.var(...)) under jit
  3. Strip out from forwarded kwargs

Example fix

// before
jnp.var(x, axis=0, out=var_buf)
// after
var_buf = jnp.var(x, axis=0)
Defensive patterns

Strategy: validation

Validate before calling

v = jnp.var(x, axis=0)  # no out kwarg

Prevention

When it happens

Trigger: jnp.var(x, out=buf), np.var(jax_array, out=...) via method delegation, or x.var(out=buf) on a JAX array.

Common situations: Statistics pipelines ported from numpy that write into preallocated buffers; kwargs-forwarding variance wrappers.

Related errors


AI-assisted analysis of jax-ml/jax@1e1c6a8fc0 (2026-08-27). Data as JSON: /api/errors/a091e3ae587b32b3. Report an issue: GitHub.

Appendix: source

Thrown at jax/_src/numpy/reductions.py:1134

     [6.22]]
  """
  if correction is None:
    correction = ddof
  elif not isinstance(ddof, int) or ddof != 0:
    raise ValueError("ddof and correction can't be provided simultaneously.")
  a = ensure_arraylike("var", a)
  return _var(a, axis=_ensure_optional_axes(axis), dtype=dtype, out=out, correction=correction, keepdims=keepdims,
              where=where, a_mean=mean)

@api.jit(static_argnames=('axis', 'dtype', 'keepdims'))
def _var(a: Array, *, axis: Axis = None, dtype: DTypeLike | None = None,
         out: None = None, correction: int | float = 0, keepdims: bool = False,
         where: ArrayLike | None = None, a_mean: ArrayLike | None = None) -> Array:
  where = check_where("var", where)
  if dtype is not None:
    dtype = dtypes.check_and_canonicalize_user_dtype(dtype, "var")
  if out is not None:
    raise NotImplementedError("The 'out' argument to jnp.var is not supported.")

  computation_dtype, dtype = _var_promote_types(a.dtype, dtype)
  a = lax.asarray(a).astype(computation_dtype)
  if a_mean is None:
    a_mean = mean(a, axis, dtype=computation_dtype, keepdims=True, where=where)
  else:
    a_mean = ensure_arraylike("var", a_mean).astype(computation_dtype)

  centered = lax.sub(a, a_mean)
  if dtypes.issubdtype(computation_dtype, np.complexfloating):
    centered = lax.real(lax.mul(centered, lax.conj(centered)))
    computation_dtype = centered.dtype  # avoid casting to complex below.
  else:
    centered = lax.square(centered)

  normalizer = _count(
      a,
      axis=axis,

View on GitHub (pinned to 1e1c6a8fc0)