What causes "Unsupported dtype for `int_from_data`" in tracel-ai/burn?

Cast the data to a supported integer dtype (e.g. i32/u32) before constructing the tensor

Unsupported dtype for `int_from_data`

Error message

Unsupported dtype for `int_from_data`

What it means

int_from_data converts host integer TensorData into a backend integer tensor. The supported set is I64/I32/I16/I8/U64/U32/U16/U8; any other dtype (floats, bool, quantized) cannot be loaded as an int tensor, so it panics with this unimplemented!.

Source

Thrown at crates/burn-cubecl/src/ops/int_tensor.rs:47

        let dtype = dtype.into();
        super::empty(shape, device, dtype)
    }

    async fn int_into_data(tensor: IntTensor<Self>) -> Result<TensorData, ExecutionError> {
        super::into_data(tensor).await
    }

    fn int_from_data(data: TensorData, device: &Device<Self>) -> IntTensor<Self> {
        match data.dtype {
            DType::I64
            | DType::I32
            | DType::I16
            | DType::I8
            | DType::U64
            | DType::U32
            | DType::U16
            | DType::U8 => super::from_data(data, device),
            _ => unimplemented!("Unsupported dtype for `int_from_data`"),
        }
    }

    fn int_to_device(tensor: IntTensor<Self>, device: &Device<Self>) -> IntTensor<Self> {
        super::to_device(tensor, device)
    }

    fn int_reshape(tensor: IntTensor<Self>, shape: Shape) -> IntTensor<Self> {
        super::reshape(tensor, shape)
    }

    fn int_slice(tensor: IntTensor<Self>, slices: &[Slice]) -> IntTensor<Self> {
        // Check if all steps are 1
        let all_steps_one = slices.iter().all(|info| info.step == 1);

        if all_steps_one {
            // Use optimized slice for step=1
            let simple_ranges: Vec<Range<usize>> = slices

View on GitHub (pinned to d16f7ba2ed)

Solutions

Cast the data to a supported integer dtype (e.g. i32/u32) before constructing the tensor
Convert floats explicitly: build the float tensor then .int() / cast to the int dtype
Verify TensorData.dtype matches the tensor type you construct (Tensor<.., Int>)
Fix data-loading code so integer arrays are stored with integer dtypes

Example fix

// before
let data = TensorData::from([1.0f32, 2.0]);
let idx = Tensor::<Backend, 1>::from_data(data, &device); // int tensor from float data -> panic
// after
let idx = Tensor::<Backend, 1>::from_data(TensorData::from([1i32, 2]), &device);

Defensive patterns

Strategy: type-guard

Validate before calling

fn int_data_supported(data: &TensorData) -> bool {
    matches!(data.dtype,
        DType::I64 | DType::I32 | DType::I16 | DType::I8 |
        DType::U64 | DType::U32 | DType::U16 | DType::U8)
}

Type guard

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

Try / catch

// Guard before creating int tensor:
if int_data_supported(&data) { Tensor::<B, D, Int>::from_data(data, &device) } else { /* cast data or build float tensor then .int() */ }

Prevention

Match TensorData.dtype to the tensor kind (Int vs Float)
Cast floats to ints explicitly before building index tensors
Validate dtypes at data-loading boundaries

When it happens

Trigger: Calling int_from_data with TensorData whose dtype is not one of the eight integer dtypes — e.g. passing float data into an int tensor constructor, or bool/quantized dtype data.

Common situations: Type confusion when building TensorData manually; passing float tensors where index/int tensors are expected; data loaded from files with a mislabeled dtype.

Related errors

AI-assisted analysis of tracel-ai/burn@d16f7ba2ed (2026-09-05). Data as JSON: /api/errors/8bc1c2b63025c310. Report an issue: GitHub.