tracel-ai/burn · error

Cannot compute min of empty tensor

Error message

Cannot compute min of empty tensor

What it means

min_view reduces a tensor view to its minimum element with reduce(); an empty iterator yields None and the expect() panics because the minimum of zero elements is undefined. Triggered by full min reductions on the ndarray backend.

Source

Thrown at crates/burn-ndarray/src/ops/base.rs:1296

            .copied()
            .reduce(|a, b| {
                if a.partial_cmp(&a).is_none() || a > b {
                    a
                } else {
                    b
                }
            })
            .expect("Cannot compute max of empty tensor");
        ArrayD::from_elem(IxDyn(&[1]), max).into_shared()
    }

    /// Min of all elements - zero-copy for borrowed storage.
    pub fn min_view(view: ArrayView<'_, E, IxDyn>) -> SharedArray<E> {
        let min = view
            .iter()
            .copied()
            .reduce(|a, b| if a < b { a } else { b })
            .expect("Cannot compute min of empty tensor");
        ArrayD::from_elem(IxDyn(&[1]), min).into_shared()
    }

    /// Min of all floating-point elements with NaN propagation.
    pub fn min_float_view(view: ArrayView<'_, E, IxDyn>) -> SharedArray<E>
    where
        E: FloatNdArrayElement,
    {
        let min = view
            .iter()
            .copied()
            .reduce(|a, b| {
                if a.partial_cmp(&a).is_none() || a < b {
                    a
                } else {
                    b
                }
            })

View on GitHub (pinned to d16f7ba2ed)

Solutions

1. Guard with a num_elements() == 0 check before calling min and handle the empty case
2. Correct upstream slicing so no dimension becomes 0
3. Skip empty tensors in metric/loss aggregation loops

Example fix

// before
let lo = empty.min(); // panics
// after
if empty.num_elements() > 0 {
    let lo = empty.min();
} else {
    // handle empty case
}

fn safe_min<E: burn_ndarray::FloatElement, const D: usize>(t: &Tensor<NdArray<E>, D>) -> Option<Tensor<NdArray<E>, 1>> {
    (t.num_elements() > 0).then(|| t.min())
}

When it happens

Trigger: Calling Tensor::min_dim / min reduction on a tensor containing zero elements - any dimension of size 0, e.g. from an empty slice range or an empty batch.

Common situations: Empty data batch reaching a min-based normalization; clipping via min on an empty intermediate tensor; dynamic/conditional code paths that produce 0-sized tensors.

Related errors

• Cannot compute max of empty tensor
• any_float: unsupported dtype {:?}
• all_float: unsupported dtype {:?}
• any_int: unsupported dtype {:?}
• all_int: unsupported dtype {:?}

AI-assisted analysis of tracel-ai/burn@d16f7ba2ed (2026-09-05). Data as JSON: /api/errors/6df93fe60b595593. Report an issue: GitHub.