jax-ml/jax · error · nb::value_error

dtype parameter is not supported by Buffer.array.

Error message

dtype parameter is not supported by Buffer.__array__.

What it means

The FFI Buffer type implements numpy's __array__ protocol but ignores the dtype parameter; passing a dtype would imply a cast that this zero-copy view cannot honor, so it is rejected.

Source

Thrown at jaxlib/ffi.cc:443

nb::tuple PyFfiAnyBuffer::DLPackDevice() const {
  return nb::make_tuple(static_cast<int32_t>(device_type_), device_ordinal_);
}

void RegisterFfiApis(nb::module_& m) {
  nb::module_ ffi_module =
      m.def_submodule("ffi", "Python bindings for the XLA FFI.");

  nb::class_<PyFfiAnyBuffer> buffer(ffi_module, "Buffer");
  buffer.def_prop_ro("dtype", xla::ValueOrThrowWrapper(&PyFfiAnyBuffer::dtype));
  buffer.def_prop_ro("ndim", &PyFfiAnyBuffer::ndim);
  buffer.def_prop_ro("shape", &PyFfiAnyBuffer::shape);
  buffer.def_prop_ro("writeable", &PyFfiAnyBuffer::writeable);
  buffer.def(
      "__array__",
      [](PyFfiAnyBuffer self, nb::object dtype, nb::object copy) {
        if (!dtype.is_none()) {
          throw nb::value_error(
              "dtype parameter is not supported by Buffer.__array__.");
        }
        if (!copy.is_none() && nb::cast<bool>(copy)) {
          throw nb::value_error(
              "Buffer.__array__ with copy=True is not supported.");
        }
        return xla::ValueOrThrow(self.NumpyArray());
      },
      nb::arg("dtype") = nb::none(), nb::arg("copy") = nb::none());
  buffer.def_prop_ro(
      "__cuda_array_interface__",
      xla::ValueOrThrowWrapper(&PyFfiAnyBuffer::CudaArrayInterface));
  buffer.def(
      "__dlpack__",
      [](PyFfiAnyBuffer self, nb::object stream, nb::object max_version,
         nb::object dl_device, nb::object copy) {
        if (!copy.is_none() && nb::cast<bool>(copy)) {
          throw nb::value_error(

View on GitHub (pinned to 1e1c6a8fc0)

Solutions

Call __array__ without dtype and cast afterwards: arr.astype(np.float32)
Do the conversion inside the kernel on the raw data pointer instead

Example fix

# before
np.asarray(buf, dtype=np.float32)
# after
np.asarray(buf).astype(np.float32)

Defensive patterns

Strategy: validation

Validate before calling

arr = buf.__array__()  # never pass dtype

Prevention

Cast after conversion with ndarray.astype
Don't write dtype-generic np.asarray(x, dtype=...) helpers that hit FFI buffers

When it happens

Trigger: np.asarray(ffi_buffer, dtype=some_dtype) or calling buffer.__array__(dtype=...) on an xla.ffi Buffer object.

Common situations: Code doing np.asarray(x, dtype=np.float32) generically on any array-like, hitting an FFI buffer during custom-call/ffi kernel testing.

Related errors

AI-assisted analysis of jax-ml/jax@1e1c6a8fc0 (2026-08-27). Data as JSON: /api/errors/32d7fd78f49e7437. Report an issue: GitHub.