huggingface/candle · error

more strides than expected in conv2d {s:?} {}

Error message

more strides than expected in conv2d {s:?} {}

What it means

candle-onnx maps the Conv `strides` attribute to one scalar for candle's conv1d/conv2d APIs; for rank-2 weights it accepts one or two equal values. A strides list with more than two elements (or otherwise unmatchable) cannot be handled, so simple_eval_ bails with this message.

Source

Thrown at candle-onnx/src/eval.rs:949

                            }
                            Some(pads) => {
                                bail!("more pads than expected in conv2d {pads:?} {}", node.name)
                            }
                        };
                        let strides = match strides {
                            None => 1,
                            Some([p]) => *p as usize,
                            Some([p1, p2]) => {
                                if p1 != p2 {
                                    bail!(
                                        "strides have to be the same on both axis {pads:?} {}",
                                        node.name
                                    )
                                }
                                *p1 as usize
                            }
                            Some(s) => {
                                bail!("more strides than expected in conv2d {s:?} {}", node.name)
                            }
                        };
                        let dilations = match dilations {
                            None => 1,
                            Some([p]) => *p as usize,
                            Some([p1, p2]) => {
                                if p1 != p2 {
                                    bail!(
                                        "dilations have to be the same on both axis {pads:?} {}",
                                        node.name
                                    )
                                }
                                *p1 as usize
                            }
                            Some(s) => {
                                bail!("more dilations than expected in conv2d {s:?} {}", node.name)
                            }
                        };

View on GitHub (pinned to d5fee525bf)

Solutions

  1. Fix the graph so the strides list length matches the weight rank (1 value for 1D, 2 for 2D)
  2. Re-export the model with the correct op (Conv1d/Conv2d) so attributes have the right arity
  3. Simplify the model with onnxsim to normalize attributes before inference
  4. Patch candle-onnx to slice or validate strides per weight rank

Example fix

// before: strides: [1, 1, 1] on a rank-2 conv
// after: strides: [1, 1]
Defensive patterns

Strategy: validation

Validate before calling

for node in &model.graph.node {
    if node.op_type == "Conv" {
        if let Some(s) = node.attribute.iter().find(|a| a.name == "strides") {
            if s.ints.len() > 2 {
                panic!("node {}: strides arity {} unsupported", node.name, s.ints.len());
            }
        }
    }
}

Type guard

fn strides_arity_ok(attr: &Attribute) -> bool {
    attr.name != "strides" || matches!(attr.ints.len(), 0..=2)
}

Try / catch

match candle_onnx::simple_eval(&model, &inputs) {
    Err(e) if e.to_string().contains("more strides than expected") => {
        eprintln!("strides list doesn't match weight rank: {e}");
    }
    other => other?,
}

Prevention

When it happens

Trigger: A Conv node whose `strides` attribute has 3+ elements (e.g. a 1D conv node exported with 3-element strides, or a corrupted/converted graph) hits the catch-all Some(s) arm.

Common situations: Graphs exported with rank mismatch between attribute length and kernel rank (e.g. 3D conv exported as Conv with rank-2 weights); automated model converters writing padded attribute lists.

Related errors

AI-assisted analysis of huggingface/candle@d5fee525bf (2026-09-02). Data as JSON: /api/errors/f8de4d61195e6fe8. Report an issue: GitHub.