tracel-ai/burn · error

a two-level scheme should have a per-tensor scale

Error message

a two-level scheme should have a per-tensor scale

What it means

Invariant check in the ndarray backend's `quantize`: the scheme is two-level (requires a global scale), but the per-tensor scale is missing from the quantization parameters, so the `global.expect(...)` panics.

Source

Thrown at crates/burn-ndarray/src/ops/qtensor.rs:156

                Some(block_size),
                QuantScheme {
                    mode: QuantMode::Symmetric,
                    #[cfg(not(feature = "export_tests"))]
                        value: QuantValue::Q8F | QuantValue::Q8S,
                    #[cfg(feature = "export_tests")]
                        value:
                        QuantValue::Q8F
                        | QuantValue::Q8S
                        | QuantValue::Q4F
                        | QuantValue::Q4S
                        | QuantValue::Q2F
                        | QuantValue::Q2S,
                    store: QuantStore::Native,
                    ..
                },
            ) => {
                let global = if global_scale_dtype(scheme).is_some() {
                    Some(global.expect("a two-level scheme should have a per-tensor scale"))
                } else {
                    None
                };
                quantize_per_block(
                    data_f.as_slice().unwrap(),
                    shape.clone(),
                    scheme,
                    block_size,
                    scales.as_slice(),
                    global,
                )
            }
            (_, scheme) => unimplemented!("Quantization not supported for scheme {scheme:?}"),
        };

        let q_bytes = QuantizedBytes {
            shape: data.shape.clone(),
            bytes: data.into_bytes(),

View on GitHub (pinned to d16f7ba2ed)

Solutions

1. Supply a per-tensor global scale in QuantizationParametersPrimitive for two-level schemes
2. Use a one-level scheme if no global scale is available
3. Check how the scheme and qparams were created (e.g., loaded from a checkpoint)

When it happens

Trigger: Thrown at crates/burn-ndarray/src/ops/qtensor.rs:156 when the library encounters an invalid state.

Common situations: See trigger scenarios.

AI-assisted analysis of tracel-ai/burn@d16f7ba2ed (2026-09-05). Data as JSON: /api/errors/7991c824fab36d46. Report an issue: GitHub.