onnx-ir 0.19.0

ONNX-IR is a pure Rust library for parsing ONNX models into an intermediate representation that can be used to generate code for various ML/DL frameworks
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122

use crate::ir::Node;

/// Configuration for LayerNorm operations
#[derive(Debug, Clone)]
pub struct LayerNormConfig {
    /// Number of features/model dimension
    pub d_model: usize,
    /// Small constant added for numerical stability
    pub epsilon: f64,
}

impl LayerNormConfig {
    /// Create a new LayerNormConfig
    pub fn new(d_model: usize) -> Self {
        Self {
            d_model,
            epsilon: 1e-5,
        }
    }

    /// Set the epsilon value
    pub fn with_epsilon(mut self, epsilon: f64) -> Self {
        self.epsilon = epsilon;
        self
    }
}

/// Create a LayerNormConfig from the attributes of the node
pub fn layer_norm_config(node: &Node) -> (LayerNormConfig, bool) {
    let weight_shape = node.inputs[1]
        .value
        .as_ref()
        .expect("LayerNorm: weight tensor must be present")
        .shape
        .clone();

    let num_features = weight_shape[0];

    // When `stash_type` is `1` (default), perform operations in 32-bit float and
    // cast the results back to original dtype
    let mut stash_type = 1;
    let mut axis = -1;
    let mut epsilon = 1e-5;

    for (key, value) in node.attrs.iter() {
        match key.as_str() {
            "axis" => axis = value.clone().into_i64(),
            "epsilon" => epsilon = value.clone().into_f32(),
            "stash_type" => stash_type = value.clone().into_i64(),
            _ => panic!("Unexpected attribute for LayerNorm: {key}"),
        }
    }

    if axis != -1 && axis != weight_shape.len() as i64 - 1 {
        panic!("LayerNorm: normalization is only supported on the last axis right now")
    }

    (
        LayerNormConfig::new(num_features).with_epsilon(epsilon as f64),
        stash_type == 1,
    )
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::ir::NodeType;
    use crate::node::test_utils::NodeBuilder;

    fn create_test_node(epsilon: f32, axis: i64, stash_type: i64, num_features: usize) -> Node {
        let weight_data = vec![1.0; num_features]; // Not important for the test
        let bias_data = vec![0.0; num_features]; // Not important for the test

        NodeBuilder::new(NodeType::LayerNormalization, "test_layernorm")
            .input_tensor_f32("X", 3, None)
            .input_tensor_f32_data("scale", weight_data, vec![num_features])
            .input_tensor_f32_data("bias", bias_data, vec![num_features])
            .output_tensor_f32("output", 3, None)
            .attr_float("epsilon", epsilon)
            .attr_int("axis", axis)
            .attr_int("stash_type", stash_type)
            .build()
    }

    #[test]
    fn test_layer_norm_config_basic() {
        let node = create_test_node(1e-5, -1, 1, 64);
        let (config, stash_type_flag) = layer_norm_config(&node);

        assert_eq!(config.d_model, 64);
        assert!(f64::abs(config.epsilon - 1e-5) < 1e-6);
        assert!(stash_type_flag);
    }

    #[test]
    fn test_layer_norm_config_no_stash_type() {
        let node = create_test_node(1e-5, -1, 0, 32);
        let (config, stash_type_flag) = layer_norm_config(&node);

        assert_eq!(config.d_model, 32);
        assert!(!stash_type_flag);
    }

    #[test]
    #[should_panic]
    fn test_layer_norm_config_invalid_axis() {
        // For a 1D weight tensor with shape [num_features],
        // both axis=0 (the first and only dim) and axis=-1 (the last dim) are valid
        // So we need to use a 2D weight tensor to test the invalid axis case

        // Create a custom node with a 2D weight tensor
        let mut node = create_test_node(1e-5, 0, 1, 64);

        // Modify the weight tensor to be 2D
        if let Some(ref mut tensor) = node.inputs[1].value {
            tensor.shape = vec![64, 64]; // Make it 2D
        }

        // Now axis=0 should trigger a panic since it's not the last dimension
        let _ = layer_norm_config(&node);
    }
}