jax-ml/jax · error · ValueError

index can't contain negative values

Error message

index can't contain negative values

What it means

jnp.pad rejects pad_width containing negative values; negative padding (numpy also disallows effective shrinking via negative widths in jnp) cannot be expressed, so any negative entry raises ValueError 'index can't contain negative values'.

Source

Thrown at jax/_src/numpy/lax_numpy.py:4191

  nd = np.ndim(array)

  if nd == 0:
    return array

  stat_funcs: dict[str, PadStatFunc] = {
      "maximum": reductions.amax,
      "minimum": reductions.amin,
      "mean": reductions.mean,
      "median": reductions.median
  }

  pad_width = _broadcast_to_pairs(pad_width, nd, "pad_width")
  pad_width_arr = np.array(pad_width)
  if pad_width_arr.shape != (nd, 2):
    raise ValueError(f"Expected pad_width to have shape {(nd, 2)}; got {pad_width_arr.shape}.")

  if np.any(pad_width_arr < 0):
    raise ValueError("index can't contain negative values")

  if mode == "constant":
    return _pad_constant(array, pad_width, asarray(constant_values))

  elif mode == "wrap":
    return _pad_wrap(array, pad_width)

  elif mode in ("symmetric", "reflect"):
    return _pad_symmetric_or_reflect(array, pad_width, str(mode), reflect_type)

  elif mode == "edge":
    return _pad_edge(array, pad_width)

  elif mode == "linear_ramp":
    end_values = _broadcast_to_pairs(end_values, nd, "end_values")
    return _pad_linear_ramp(array, pad_width, end_values)

  elif mode in stat_funcs:

View on GitHub (pinned to 1e1c6a8fc0)

Solutions

1. Clamp computed widths: max(0, target - x.shape[0])
2. Use dynamic_slice or lax.slice to actually crop when shrink is intended

Example fix

// before
jnp.pad(x, (target - x.shape[0], 0))
// after
jnp.pad(x, (max(0, target - x.shape[0]), 0))

assert np.all(np.asarray(pad_width) >= 0), 'negative pad width'

When it happens

Trigger: jnp.pad(x, (-1, 2)) or computed widths that go negative, e.g. (target_len - x.shape[0]) when the array is already longer than target.

Common situations: Padding sequences to a fixed length where some inputs already exceed the target, making the computed pad amount negative.

Related errors

• `{name}` entries must be the same shape: {nvals}
• jnp.pad: {name} with {nd=} has unsupported shape {nvals.shap
• Cannot apply '{}' padding to empty axis
• jnp.pad: constant_values has unsupported shape {constant_val
• Shape polymorphism is supported for jnp.pad with 'reflect' o

AI-assisted analysis of jax-ml/jax@1e1c6a8fc0 (2026-08-27). Data as JSON: /api/errors/e569a093b7f7dab5. Report an issue: GitHub.