tracel-ai/burn · error

interpolate: unsupported mode {:?} / dtype {:?}

Error message

interpolate: unsupported mode {:?} / dtype {:?}

What it means

The Flex backend's interpolate op dispatches on the (InterpolateMode, dtype) pair. Every combination of Nearest/Bilinear/Bicubic/Lanczos3 with F32/F64/F16/BF16 is implemented; any other pair — meaning a non-float dtype, since all four InterpolateMode variants are covered — hits the catch-all arm and panics naming both the mode and dtype.

Source

Thrown at crates/burn-flex/src/ops/module.rs:603

            (InterpolateMode::Bicubic, DType::F16) => {
                interpolate::interpolate_bicubic_f16(x, output_size, options.align_corners)
            }
            (InterpolateMode::Bicubic, DType::BF16) => {
                interpolate::interpolate_bicubic_bf16(x, output_size, options.align_corners)
            }
            (InterpolateMode::Lanczos3, DType::F32) => {
                interpolate::interpolate_lanczos3_f32(x, output_size, options.align_corners)
            }
            (InterpolateMode::Lanczos3, DType::F64) => {
                interpolate::interpolate_lanczos3_f64(x, output_size, options.align_corners)
            }
            (InterpolateMode::Lanczos3, DType::F16) => {
                interpolate::interpolate_lanczos3_f16(x, output_size, options.align_corners)
            }
            (InterpolateMode::Lanczos3, DType::BF16) => {
                interpolate::interpolate_lanczos3_bf16(x, output_size, options.align_corners)
            }
            (mode, dtype) => panic!(
                "interpolate: unsupported mode {:?} / dtype {:?}",
                mode, dtype
            ),
        }
    }

    fn interpolate_backward(
        x: FloatTensor<Flex>,
        grad: FloatTensor<Flex>,
        output_size: [usize; 2],
        options: InterpolateOptions,
    ) -> FloatTensor<Flex> {
        match (options.mode, x.dtype()) {
            (InterpolateMode::Nearest, DType::F32) => {
                interpolate::interpolate_nearest_backward_f32(
                    x,
                    grad,
                    output_size,

View on GitHub (pinned to d16f7ba2ed)

Solutions

1. Cast the input to a supported float dtype before interpolating: x.cast(DType::F32).
2. Check the reported dtype in the panic message and fix the upstream tensor producer.
3. Add match arms for the missing (mode, dtype) combination calling interpolate::<mode>_<dtype> in crates/burn-flex/src/ops/module.rs.

Example fix

// before
let up = img_u8.interpolate(Bilinear, [256, 256]); // panics: (Bilinear, U8)
// after
let up = img_u8.cast(DType::F32).interpolate(Bilinear, [256, 256]);

assert!(matches!(x.dtype(), burn::tensor::DType::F32 | burn::tensor::DType::F64 | burn::tensor::DType::F16 | burn::tensor::DType::BF16), "interpolate needs a float tensor, got {:?} (mode {:?})", x.dtype(), options.mode);

fn is_float_dtype(d: burn::tensor::DType) -> bool {
    matches!(d, burn::tensor::DType::F32 | burn::tensor::DType::F64 | burn::tensor::DType::F16 | burn::tensor::DType::BF16)
}

// Convert u8/int images before interpolating:
let x = if is_float_dtype(x.dtype()) { x } else { x.cast(burn::tensor::DType::F32) };
let up = x.interpolate(burn::tensor::module::InterpolateMode::Bilinear, [256, 256]);

When it happens

Trigger: Calling Tensor::interpolate (or Upsample layers) on the Flex backend with an input tensor whose dtype is not F32/F64/F16/BF16, regardless of mode; the message reports e.g. (Bilinear, U8) or (Bilinear, I32).

Common situations: Integer image tensors (u8 pixel data) passed to upsample without conversion; forgetting .float() after loading images; dtype drift from a quantized pipeline or checkpoint load.

Related errors

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

AI-assisted analysis of tracel-ai/burn@d16f7ba2ed (2026-09-05). Data as JSON: /api/errors/90ef4f4a93faa763. Report an issue: GitHub.