tracel-ai/burn · error

int_select_assign: unsupported dtype {:?}

Error message

int_select_assign: unsupported dtype {:?}

What it means

int_select_assign panics on the set-assignment path when the tensor dtype is not one of the eight implemented integer widths. select_assign writes fixed-width values at index positions, so unknown dtypes cannot be written safely and the library panics instead.

Source

Thrown at crates/burn-flex/src/ops/int.rs:341

                    DType::I16 => crate::ops::gather_scatter::select_assign::<i16>(
                        tensor, dim, indices, value,
                    ),
                    DType::I8 => {
                        crate::ops::gather_scatter::select_assign::<i8>(tensor, dim, indices, value)
                    }
                    DType::U64 => crate::ops::gather_scatter::select_assign::<u64>(
                        tensor, dim, indices, value,
                    ),
                    DType::U32 => crate::ops::gather_scatter::select_assign::<u32>(
                        tensor, dim, indices, value,
                    ),
                    DType::U16 => crate::ops::gather_scatter::select_assign::<u16>(
                        tensor, dim, indices, value,
                    ),
                    DType::U8 => {
                        crate::ops::gather_scatter::select_assign::<u8>(tensor, dim, indices, value)
                    }
                    dt => panic!("int_select_assign: unsupported dtype {:?}", dt),
                }
            }
            burn_backend::tensor::IndexingUpdateOp::Add => {
                debug_assert_eq!(
                    tensor.dtype(),
                    value.dtype(),
                    "int_select_assign: dtype mismatch"
                );
                match tensor.dtype() {
                    DType::I64 => {
                        crate::ops::gather_scatter::select_add::<i64>(tensor, dim, indices, value)
                    }
                    DType::I32 => {
                        crate::ops::gather_scatter::select_add::<i32>(tensor, dim, indices, value)
                    }
                    DType::I16 => {
                        crate::ops::gather_scatter::select_add::<i16>(tensor, dim, indices, value)
                    }

View on GitHub (pinned to d16f7ba2ed)

Solutions

1. Cast tensor and value to a supported int dtype before the select_assign call
2. Guarantee tensor.dtype() == value.dtype() as a precondition
3. Confirm Set is the intended update op; this panic is the Set arm
4. Add a select_assign match arm calling crate::ops::gather_scatter::select_assign::<T> for new dtypes

Example fix

// before
tensor.select_assign(dim, indices, value, IndexingUpdateOp::Set); // tensor is DType::Bool
// after
let tensor = tensor.cast(DType::I32);
let value = value.cast(DType::I32);
tensor.select_assign(dim, indices, value, IndexingUpdateOp::Set);

assert_eq!(tensor.dtype(), value.dtype(), "select_assign: dtype mismatch");
assert!(is_supported_int_dtype(tensor.dtype()), "unsupported dtype {:?}", tensor.dtype());

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

When it happens

Trigger: Calling int_select_assign with IndexingUpdateOp::Set on a tensor whose dtype is not i64/i32/i16/i8/u64/u32/u16/u8, or when tensor and value dtypes differ so the wrong arm is evaluated.

Common situations: Assigning into selected ranges of bool-masked tensors, dtype drift between target and value, or a burn upgrade introducing a DType variant burn-flex does not cover.

Related errors

• float_select_assign: unsupported dtype {:?}
• int_gather: unsupported dtype {:?}
• int_scatter: unsupported dtype {:?}
• int_scatter_nd: unsupported dtype {:?}
• int_gather_nd: unsupported dtype {:?}

AI-assisted analysis of tracel-ai/burn@d16f7ba2ed (2026-09-05). Data as JSON: /api/errors/e5490af705d25389. Report an issue: GitHub.