burn-onnx 0.21.0-pre.3

Library for importing ONNX models into the Burn framework
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
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140

use super::prelude::*;
use burn_store::TensorSnapshot;

impl NodeCodegen for onnx_ir::node::conv_transpose3d::ConvTranspose3dNode {
    fn inputs(&self) -> &[Argument] {
        // Filter inputs only dynamic and constant
        &self.inputs
    }

    fn outputs(&self) -> &[Argument] {
        &self.outputs
    }

    fn field(&self) -> Option<Field> {
        let name = Ident::new(&self.name, Span::call_site());
        let weight_shape = self.inputs[1]
            .ty
            .static_shape_known()
            .expect("ConvTranspose3d: weight tensor shape must be known at codegen time");
        let groups = self.config.groups;
        let channels = [weight_shape[0], weight_shape[1] * groups].to_tokens();
        let kernel_size = self.config.kernel_size.to_tokens();
        let stride = self.config.stride.to_tokens();
        let dilation = self.config.dilation.to_tokens();
        let groups = groups.to_tokens();
        let padding = self.config.padding.to_tokens();
        let padding_out = self.config.padding_out.to_tokens();
        let bias = self.inputs.len() == 3;

        Some(Field::new(
            self.name.clone(),
            quote! {
                ConvTranspose3d<B>
            },
            quote! {
                let #name = ConvTranspose3dConfig::new(#channels, #kernel_size)
                    .with_stride(#stride)
                    .with_padding(#padding)
                    .with_padding_out(#padding_out)
                    .with_dilation(#dilation)
                    .with_groups(#groups)
                    .with_bias(#bias)
                    .init(device);
            },
        ))
    }

    fn forward(&self, scope: &mut ScopeAtPosition<'_>) -> TokenStream {
        let input = scope.arg(self.inputs.first().unwrap());
        let output = arg_to_ident(self.outputs.first().unwrap());
        let field = Ident::new(&self.name, Span::call_site());

        quote! {
            let #output = self.#field.forward(#input);
        }
    }
    fn register_imports(&self, imports: &mut BurnImports) {
        imports.register("burn::nn::conv::ConvTranspose3d");
        imports.register("burn::nn::conv::ConvTranspose3dConfig");
    }

    fn collect_snapshots(&self, field_name: &str) -> Vec<TensorSnapshot> {
        use crate::burn::node_traits::create_lazy_snapshot;

        let mut snapshots = vec![];

        // Weight tensor (input index 1)
        // ONNX ConvTranspose weight: [in_channels, out_channels/groups, kD, kH, kW]
        // Burn ConvTranspose3d weight: [channels_in, channels_out/groups, kD, kH, kW]
        // These layouts match! No transformation needed.
        if let Some(weight_input) = self.inputs.get(1) {
            let weight_path = format!("{}.weight", field_name);
            if let Some(snapshot) =
                create_lazy_snapshot(weight_input, &weight_path, "ConvTranspose3d")
            {
                snapshots.push(snapshot);
            }
        }

        // Bias tensor (input index 2, optional)
        if self.inputs.len() > 2
            && let Some(bias_input) = self.inputs.get(2)
        {
            let bias_path = format!("{}.bias", field_name);
            if let Some(snapshot) = create_lazy_snapshot(bias_input, &bias_path, "ConvTranspose3d")
            {
                snapshots.push(snapshot);
            }
        }

        snapshots
    }
}

#[cfg(test)]
mod tests {
    use super::super::test_helpers::*;
    use burn::tensor::DType;
    use insta::assert_snapshot;
    use onnx_ir::node::conv_transpose3d::{
        ConvTranspose3dConfig, ConvTranspose3dNode, ConvTranspose3dNodeBuilder,
    };

    fn create_conv_transpose_3d_node(name: &str) -> ConvTranspose3dNode {
        let config =
            ConvTranspose3dConfig::new([3, 3, 3], [1, 1, 1], [1, 1, 1], [1, 1, 1], [0, 0, 0], 1);

        ConvTranspose3dNodeBuilder::new(name)
            .input_tensor("input", 5, DType::F32)
            .input_static_tensor_shape("weight", vec![3, 64, 3, 3, 3], DType::F32)
            .input_static_tensor_shape("bias", vec![64], DType::F32)
            .output_tensor("output", 5, DType::F32)
            .config(config)
            .build()
    }

    #[test]
    fn test_conv_transpose_3d_forward() {
        let node = create_conv_transpose_3d_node("conv_transpose1");
        let code = codegen_forward_default(&node);
        assert_snapshot!(code, @r"
        pub fn forward(&self, input: Tensor<B, 5>) -> Tensor<B, 5> {
            let output = self.conv_transpose1.forward(input);
            output
        }
        ");
    }

    #[test]
    fn test_conv_transpose_3d_forward_with_clone() {
        let node = create_conv_transpose_3d_node("conv_transpose1");
        let code = codegen_forward_with_clone(&node);
        assert_snapshot!(code, @r"
        pub fn forward(&self, input: Tensor<B, 5>) -> Tensor<B, 5> {
            let output = self.conv_transpose1.forward(input.clone());
            output
        }
        ");
    }
}