jax-ml/jax · error · ValueError

Need at least one array to stack.

Error message

Need at least one array to stack.

What it means

jnp.stack must join at least one array along the new axis to determine the output shape. It fires when the arrays sequence is empty, leaving no elements to stack.

Solutions

  1. Guard with `if not arrays: ...` and supply a default/zero-shaped result
  2. Fix upstream logic so the list is never empty

Example fix

// before
jnp.stack(chunks, axis=0)
// after
if not chunks:
    raise ValueError('no chunks to stack')
jnp.stack(chunks, axis=0)
Defensive patterns

Strategy: validation

Validate before calling

if not arrays:
    raise ValueError('nothing to stack')

Type guard

def non_empty(seq) -> bool:
    return len(seq) > 0

Prevention

When it happens

Trigger: jnp.stack([]) or jnp.stack([], axis=1); also jnp.stack of an empty generator/list built by a comprehension that filters everything out.

Common situations: Dynamic batch accumulation where a filter or conditions yield an empty list; empty minibatches; empty results in data pipelines.

Related errors


AI-assisted analysis of jax-ml/jax@1e1c6a8fc0 (2026-08-27). Data as JSON: /api/errors/ea917e2f2df281f0. Report an issue: GitHub.

Appendix: source

Thrown at jax/_src/numpy/lax_numpy.py:4429

    >>> jnp.stack([x, y])
    Array([[1, 2, 3],
           [4, 5, 6]], dtype=int32)
    >>> jnp.stack([x, y], axis=1)
    Array([[1, 4],
           [2, 5],
           [3, 6]], dtype=int32)

    :func:`~jax.numpy.unstack` performs the inverse operation:

    >>> arr = jnp.stack([x, y], axis=1)
    >>> x, y = jnp.unstack(arr, axis=1)
    >>> x
    Array([1, 2, 3], dtype=int32)
    >>> y
    Array([4, 5, 6], dtype=int32)
  """
  if not len(arrays):
    raise ValueError("Need at least one array to stack.")
  if out is not None:
    raise NotImplementedError("The 'out' argument to jnp.stack is not supported.")
  if isinstance(arrays, (np.ndarray, Array)):
    axis = _canonicalize_axis(axis, arrays.ndim)
    return concatenate(expand_dims(arrays, axis + 1), axis=axis, dtype=dtype)
  else:
    arrays = util.ensure_arraylike_tuple("stack", arrays)
    if dtype is not None:
      arrays = [asarray(a, dtype=dtype) for a in arrays]
    else:
      arrays = util.promote_dtypes(*arrays)
    return lax.stack(arrays, axis=axis)


@export
@api.jit(static_argnames="axis", inline=True)
def unstack(x: ArrayLike, /, *, axis: int = 0) -> tuple[Array, ...]:
  """Unstack an array along an axis.

View on GitHub (pinned to 1e1c6a8fc0)