tracel-ai/burn · error

Can't format yet

Error message

Can't format yet

What it means

TensorData's Display implementation (crates/burn-std/src/data/tensor/base.rs:512-563) formats quantized tensors by casting values to i8 and printing them. Only symmetric quantization with Q8/Q4/Q2 QuantValues is handled; E4M3, E5M2, and E2M1 float-point quantized data hits an explicit `unimplemented!` because formatting for float-point quantized element types has not been written. This is a deliberate placeholder panic, not a data corruption or logic bug.

Source

Thrown at crates/burn-std/src/data/tensor/base.rs:551

                    value:
                        QuantValue::Q8F
                        | QuantValue::Q8S
                        // Display sub-byte values as i8
                        | QuantValue::Q4F
                        | QuantValue::Q4S
                        | QuantValue::Q2F
                        | QuantValue::Q2S,
                    ..
                } => {
                    format!("{:?} {scheme:?}", self.iter::<i8>().collect::<Vec<_>>())
                },
                QuantScheme {
                        mode: QuantMode::Symmetric,
                        value:
                            QuantValue::E4M3 | QuantValue::E5M2 | QuantValue::E2M1,
                        ..
                    } => {
                        unimplemented!("Can't format yet");
                    }
                QuantScheme {
                    mode: QuantMode::Lookup,
                    ..
                } => {
                    format!("<lookup-quantized> {scheme:?}")
                }
            },
        };
        f.write_str(fmt.as_str())
    }
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::*;
    use ::rand::{

View on GitHub (pinned to d16f7ba2ed)

Solutions

1. Do not rely on Display for E4M3/E5M2/E2M1 quantized tensors; use the Debug impl (`{:?}` on TensorData) which prints raw bytes and dtype without this match arm.
2. Inspect values via dequantization APIs instead of Display (convert to float then print).
3. Track/upgrade burn to a version where E4M3/E5M2/E2M1 Display formatting is implemented.
4. If printing is essential, manually unpack bytes and format them yourself.

Example fix

// before
println!("{}", quantized_tensor_data); // panics: Can't format yet
// after
println!("{:?}", quantized_tensor_data); // or dequantize first, then print floats

// before formatting a quantized tensor
fn is_displayable_quantized(dtype: &DType) -> bool {
    match dtype {
        DType::QFloat(s) => !matches!(
            s,
            QuantScheme { mode: QuantMode::Symmetric, value: QuantValue::E4M3 | QuantValue::E5M2 | QuantValue::E2M1, .. }
        ),
        _ => true,
    }
}
if !is_displayable_quantized(&data.dtype) { eprintln!("{:?}", data); } else { println!("{}", data); }

fn supports_q_display(scheme: &QuantScheme) -> bool {
    !matches!(
        scheme,
        QuantScheme { mode: QuantMode::Symmetric, value: QuantValue::E4M3 | QuantValue::E5M2 | QuantValue::E2M1, .. }
    )
}

// Display panics rather than returning Result; guard before printing
if supports_q_display(&scheme) {
    println!("{}", data);
} else {
    println!("{:?}", data); // Debug fallback
}

When it happens

Trigger: Calling `format!("{}", tensor_data)`, `println!("{}", ...)` or any Display usage (e.g. `to_string()`) on a TensorData whose dtype is DType::QFloat(scheme) where scheme.mode is Symmetric and scheme.value is QuantValue::E4M3, E5M2, or E2M1.

Common situations: Debug-logging a float8/float4-quantized tensor (e.g. after quantize() with q_float E4M3 scheme) or printing tensor contents in a REPL/test while working with FP8 quantization in burn.

Related errors

• todo!("Quantization not supported yet")
• unimplemented!()
• Not yet implemented for iteration
• lookup quantization is not supported for iteration
• Not yet supported

AI-assisted analysis of tracel-ai/burn@d16f7ba2ed (2026-09-05). Data as JSON: /api/errors/504c76769668b717. Report an issue: GitHub.