tracel-ai/burn · error

Unsupported dtype: {other:?}

Error message

Unsupported dtype: {other:?}

What it means

int_abs must dispatch the abs computation to the concrete integer type stored in the tensor. If the tensor's dtype is not one of the supported signed integer types (i64/i32/i16/i8) and not an unsigned type (returned as-is), the dtype is unexpected and the backend panics with 'Unsupported dtype'.

Source

Thrown at crates/burn-ndarray/src/ops/int_tensor.rs:400

    fn int_clamp(tensor: NdArrayTensor, min: Scalar, max: Scalar) -> NdArrayTensor {
        execute_with_int_dtype!(tensor, |array| NdArrayMathOps::clamp(
            array,
            min.elem(),
            max.elem()
        ))
    }

    fn int_abs(tensor: NdArrayTensor) -> NdArrayTensor {
        match tensor.dtype() {
            DType::I64 | DType::I32 | DType::I16 | DType::I8 => {
                execute_with_dtype!(tensor, I, NdArrayMathOps::abs, [
                    I64 => i64, I32 => i32, I16 => i16, I8 => i8
                ])
            }
            // Already unsigned
            DType::U64 | DType::U32 | DType::U16 | DType::U8 => tensor,
            other => panic!("Unsupported dtype: {other:?}"),
        }
    }

    fn int_into_float(tensor: NdArrayTensor, out_dtype: FloatDType) -> FloatTensor<Self> {
        execute_with_float_out_dtype!(out_dtype, F, {
            execute_with_int_dtype!(tensor, IntElem, |array: SharedArray<IntElem>| {
                array.mapv(|a: IntElem| a.elem::<F>()).into_shared()
            })
        })
    }

    fn int_swap_dims(tensor: NdArrayTensor, dim1: usize, dim2: usize) -> NdArrayTensor {
        execute_with_int_dtype!(tensor, |array| NdArrayOps::swap_dims(array, dim1, dim2))
    }

    fn int_random(
        shape: Shape,
        distribution: Distribution,

View on GitHub (pinned to d16f7ba2ed)

Solutions

1. Ensure the int tensor uses a supported dtype (I64/I32/I16/I8 or unsigned, which is returned unchanged).
2. Cast the tensor to a supported int dtype (e.g. .int() / int cast to i32) before abs.
3. Upgrade or downgrade burn so the backend and core DType enum versions match.
4. Convert to float, take abs, and convert back as a workaround.
5. Report the missing dtype case to the burn repository if a legit variant is unhandled.

Example fix

// before
let t: Tensor<_, Int, NdArray> = ...; // dtype not in supported set
let a = t.abs();
// after
let t = t.cast(burn::tensor::DType::I32); // supported signed dtype
let a = t.abs();

assert!(matches!(t.dtype(), DType::I64 | DType::I32 | DType::I16 | DType::I8 | DType::U64 | DType::U32 | DType::U16 | DType::U8), "abs: unsupported int dtype");

fn abs_supported(dt: DType) -> bool {
    matches!(dt, DType::I64 | DType::I32 | DType::I16 | DType::I8
        | DType::U64 | DType::U32 | DType::U16 | DType::U8)
}

When it happens

Trigger: Calling Tensor::abs on an Int tensor whose dtype falls outside the I64/I32/I16/I8/U64/U32/U16/U8 set — typically only possible with a mismatched/foreign dtype variant or an enum extended in a newer burn version without updating this match.

Common situations: Using a burn version where DType gained new variants while burn-ndarray's int_abs wasn't updated; dtype confusion after converting tensors across backends; bool tensors mis-typed as int.

Related errors

• burn-flex does not support Bool(U32) storage (only Native an
• compare_int: unsupported dtype {:?}
• compare_int_elem: unsupported dtype {:?}
• any_float: unsupported dtype {:?}
• all_float: unsupported dtype {:?}

AI-assisted analysis of tracel-ai/burn@d16f7ba2ed (2026-09-05). Data as JSON: /api/errors/56b74d53c0123f48. Report an issue: GitHub.