huggingface/candle · error

cannot find 'value' attr in 'Constant' for {}

Error message

cannot find 'value' attr in 'Constant' for {}

What it means

The ONNX Constant operator can carry its output in several attributes (value, value_float, value_ints, sparse_value, etc.). candle-onnx only implements the tensor-valued `value` attribute; when a Constant node has none it bails with this message naming the node. The code comments that sparse_value etc. are not yet supported.

Source

Thrown at candle-onnx/src/eval.rs:1098

                let output = PReLU::new(slope.clone(), false).forward(input)?;
                values.insert(node.output[0].clone(), output);
            }
            "Ceil" => {
                let input = get(&node.input[0])?;
                let output = input.ceil()?;
                values.insert(node.output[0].clone(), output);
            }
            "Floor" => {
                let input = get(&node.input[0])?;
                let output = input.floor()?;
                values.insert(node.output[0].clone(), output);
            }
            // https://github.com/onnx/onnx/blob/main/docs/Operators.md#Constant
            "Constant" => {
                let value = match node.attribute.iter().find(|attr| attr.name == "value") {
                    None => {
                        // TODO: support sparse_value etc.
                        bail!("cannot find 'value' attr in 'Constant' for {}", node.name)
                    }
                    Some(value) => value,
                };
                let output = match value.r#type() {
                    AttributeType::Tensor => {
                        let t = value.t.as_ref().unwrap();
                        get_tensor(t, &node.name)?
                    }
                    rtype => bail!("unsupported 'value' type {rtype:?} for {}", node.name),
                };

                values.insert(node.output[0].clone(), output);
            }
            // https://github.com/onnx/onnx/blob/main/docs/Operators.md#Cast
            "Cast" => {
                let input = get(&node.input[0])?;
                let dt: i64 = *get_attr(node, "to")?;
                let dtype = match DataType::try_from(dt as i32) {

View on GitHub (pinned to d5fee525bf)

Solutions

  1. Convert scalar/int Constant nodes to tensor `value` attributes in the graph (onnx.helper.make_tensor) before inference
  2. Run an ONNX graph rewrite/simplifier (onnxsim) which often folds Constant nodes away
  3. Replace Constant nodes with equivalent Initializers in the model
  4. Patch candle-onnx to support value_float/value_ints/sparse_value attributes

Example fix

// before: onnx.helper.make_node('Constant', [], ['k'], value_float=1.0)
// after
node = onnx.helper.make_node('Constant', [], ['k'])
node.attribute.append(onnx.helper.make_attribute(
    'value', numpy_helper.from_array(np.array([1.0], dtype=np.float32), name='k')))
Defensive patterns

Strategy: validation

Validate before calling

for node in &model.graph.node {
    if node.op_type == "Constant" && !node.attribute.iter().any(|a| a.name == "value") {
        panic!("Constant node '{}' lacks a 'value' attribute", node.name);
    }
}

Type guard

fn has_value_attr(node: &NodeProto) -> bool {
    node.attribute.iter().any(|a| a.name == "value")
}

Try / catch

match candle_onnx::simple_eval(&model, &inputs) {
    Err(e) if e.to_string().contains("cannot find 'value' attr in 'Constant'") => {
        eprintln!("convert scalar Constants to tensor values: {e}");
    }
    other => other?,
}

Prevention

When it happens

Trigger: simple_eval encounters a Constant node whose value is expressed via value_float, value_ints, value_int64, sparse_value, or is missing entirely — common with scalar constants.

Common situations: ONNX models exported by frameworks that emit scalar Constants as value_float/value_ints instead of tensor value; models using sparse constants; graphs produced by older or newer exporters than candle-onnx supports.

Related errors

AI-assisted analysis of huggingface/candle@d5fee525bf (2026-09-02). Data as JSON: /api/errors/f9bca01c825c828c. Report an issue: GitHub.