huggingface/candle · error

no graph defined in proto

Error message

no graph defined in proto

What it means

simple_eval evaluates an ONNX ModelProto directly. The message is thrown when the ModelProto has no `graph` field set (model.graph is None). An ONNX model without a graph carries no computation and cannot be executed.

Source

Thrown at candle-onnx/src/eval.rs:244

            }
        },
        Err(_) => {
            bail!("unsupported 'value' data-type {} for {name}", t.data_type,)
        }
    }
}

// This function provides a direct evaluation of the proto.
// Longer-term, we should first convert the proto to an intermediate representation of the compute
// graph so as to make multiple evaluations more efficient.
// An example upside of this would be to remove intermediary values when they are not needed
// anymore.
pub fn simple_eval(
    model: &onnx::ModelProto,
    mut inputs: HashMap<String, Value>,
) -> Result<HashMap<String, Value>> {
    let graph = match &model.graph {
        None => bail!("no graph defined in proto"),
        Some(graph) => graph,
    };
    simple_eval_(graph, &mut inputs)
}

fn simple_eval_(
    graph: &onnx::GraphProto,
    values: &mut HashMap<String, Value>,
) -> Result<HashMap<String, Value>> {
    for t in graph.initializer.iter() {
        let tensor = get_tensor(t, t.name.as_str())?;
        values.insert(t.name.to_string(), tensor);
    }
    for input in graph.input.iter() {
        let input_type = match &input.r#type {
            Some(input_type) => input_type,
            None => continue,
        };

View on GitHub (pinned to d5fee525bf)

Solutions

1. Verify the file is a real ONNX model (starts with protobuf serialization of ModelProto; check with onnx.load and model.graph).
2. If constructing ModelProto manually, set model.graph before calling simple_eval.
3. Re-export or re-download the model; the source file is truncated or corrupt.
4. Wrap eval in error handling and surface a clear 'invalid/corrupt ONNX model' message to users.

Example fix

// before
let out = simple_eval(&model, inputs)?;
// after
assert!(model.graph.is_some(), "not a valid onnx model");
let out = simple_eval(&model, inputs)?;

fn has_graph(model: &onnx::ModelProto) -> bool { model.graph.is_some() }
// call before simple_eval

fn is_executable_model(model: &onnx::ModelProto) -> bool {
    model.graph.as_ref().map_or(false, |g| !g.node.is_empty())
}

let out = simple_eval(&model, inputs).map_err(|e| {
    if e.to_string().contains("no graph defined") { anyhow!("invalid or corrupt ONNX model file") } else { e }
})?;

When it happens

Trigger: Calling candle_onnx::simple_eval(&model, inputs) with a deserialized ModelProto whose graph field was never set — e.g. a hand-built proto, a proto loaded from a non-model file, or a stripped/corrupted .onnx file.

Common situations: Pointing the loader at the wrong file (metadata rather than the model); constructing ModelProto programmatically and forgetting to set graph; protobuf decoding succeeding but yielding an empty message.

Related errors

• unsupported 'value' data-type {} for {name}
• unsupported input type {type_:?}
• attribute {} was of type TENSOR, but no tensor was found
• attribute {} of type TENSOR was an invalid data_type number
• attribute {} of type TENSOR has an unsupported data_type {}

AI-assisted analysis of huggingface/candle@d5fee525bf (2026-09-02). Data as JSON: /api/errors/013db115d66324a5. Report an issue: GitHub.