What causes "Can read the data without error" in tracel-ai/burn?

Check the tensor's device is reachable and healthy before argwhere (device logs, connection state for remote backends).

Can read the data without error

Error message

Can read the data without error

What it means

This panic occurs in bool_argwhere after awaiting B::bool_into_data(tensor), which materializes a boolean tensor's data on the host to count nonzero elements for the argwhere (indices of true values) operation. The expect fires if the backend's data-read future returns an error — i.e. the tensor data could not be transferred/computed into a readable Data struct.

Solutions

Check the tensor's device is reachable and healthy before argwhere (device logs, connection state for remote backends).
Verify the tensor handle is still valid — not consumed or dropped elsewhere in the graph.
Look for a root-cause backend error (CUDA OOM, driver reset) reported before this expect fires.
For remote/communication backends, re-establish the peer connection and retry the operation.

Example fix

// before
let data = B::bool_into_data(tensor).await.expect("Can read the data without error");
// after
let data = B::bool_into_data(tensor).await
    .context("failed to read bool tensor data in argwhere; check device health")?;

Defensive patterns

Strategy: try-catch

Try / catch

let data = B::bool_into_data(tensor).await
    .map_err(|e| anyhow::anyhow!("argwhere: bool_into_data failed: {e}"))?;

Prevention

Confirm the tensor device is healthy and connected before data reads.
Avoid reusing tensor handles after ownership transfer into other ops.
Watch GPU memory/driver logs; OOM often surfaces at into_data.

When it happens

Trigger: The underlying backend read fails during bool_into_data — e.g. the tensor handle is invalid (freed or created on a device that errored), a backend kernel fails during the async read, or a remote/communication-backed backend loses its peer connection mid-transfer.

Common situations: Calling argwhere on a tensor whose device connection dropped (distributed/WebSocket tensor transfer failed); using a tensor that was already consumed/freed by another op; GPU driver or OOM errors surfacing through the into_data read.

Related errors

AI-assisted analysis of tracel-ai/burn@d16f7ba2ed (2026-09-05). Data as JSON: /api/errors/905c86eb81c5d95b. Report an issue: GitHub.

Appendix: source

Thrown at crates/burn-backend/src/backend/ops/bool_tensor.rs:579

    ///
    /// * `tensor` - The input tensor.
    /// * `out_dtype` - The output tensor dtype.
    ///
    /// # Returns
    ///
    /// A 2D tensor containing the indices of all non-zero elements of the given tensor.
    /// Each row contains the indices of a non-zero element.
    fn bool_argwhere(
        tensor: BoolTensor<B>,
        out_dtype: IntDType,
    ) -> impl Future<Output = IntTensor<B>> + 'static + Send {
        async move {
            // Size of each output tensor is variable (= number of nonzero elements in the tensor).
            // Reading the data to count the number of truth values might cause sync but is required.
            let device = &tensor.device();
            let data = B::bool_into_data(tensor)
                .await
                .expect("Can read the data without error");
            argwhere_data::<B>(data, device, out_dtype)
        }
    }

    /// Broadcasts the bool `tensor` to the given `shape`.
    fn bool_expand(tensor: BoolTensor<B>, shape: Shape) -> BoolTensor<B>;

    /// Unfold windows along a dimension.
    ///
    /// Returns a view of the tensor with all complete windows of size `size` in dimension `dim`;
    /// where windows are advanced by `step` at each index.
    ///
    /// The number of windows is `0` when `shape[dim] < size`, and otherwise
    /// `(shape[dim] - size) / step + 1`.
    ///
    /// # Arguments
    ///
    /// * `tensor` - The input tensor to unfold; of shape ``[pre=..., dim shape, post=...]``

View on GitHub (pinned to d16f7ba2ed)