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
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202

use super::prelude::*;

impl NodeCodegen for onnx_ir::node::det::DetNode {
    fn inputs(&self) -> &[Argument] {
        &self.inputs
    }

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

    fn forward(&self, scope: &mut ScopeAtPosition<'_>) -> TokenStream {
        let input_arg = self.inputs.first().unwrap();
        let output_arg = self.outputs.first().unwrap();
        let input = scope.arg(input_arg);
        let output = arg_to_ident(output_arg);

        let input_rank = match &input_arg.ty {
            ArgType::Tensor(t) => t.rank,
            other => unreachable!("Det input type validated in onnx-ir, got {other:?}"),
        };

        match input_rank {
            2 => {
                // 2D non-batched: compute determinant via LU decomposition with partial pivoting
                // det(A) = sign(P) * product_of_diagonal(U), where PA = LU
                quote! {
                    let #output = {
                        let (lu_mat, permutations) =
                            burn::tensor::linalg::lu_decomposition(#input);
                        let det_u =
                            burn::tensor::linalg::diag::<B, 2usize, 1usize, Float>(lu_mat)
                                .prod();
                        let n = permutations.dims()[0];
                        let perm_f = permutations.float().cast(burn::tensor::DType::F32);
                        let perm_col = perm_f.clone().reshape([n, 1]);
                        let perm_row = perm_f.reshape([1, n]);
                        let inv_count = (perm_row - perm_col)
                            .triu(1i64)
                            .lower_elem(0.0f32)
                            .int()
                            .cast(burn::tensor::DType::I64)
                            .sum();
                        let parity = inv_count.remainder_scalar(2i64);
                        let device = det_u.device();
                        let dtype = det_u.dtype();
                        let sign = Tensor::<B, 1, Float>::from_data(
                            [1.0f32, -1.0f32],
                            (&device, dtype,)
                        )
                        .select(0, parity);
                        det_u * sign
                    };
                }
            }
            _ => {
                // Batched inputs: loop over all batch dimensions using LU decomposition
                // per-matrix. This is a fallback; a native batched det in Burn would be
                // more efficient.
                let batch_rank = input_rank - 2;
                let batch_dims: Vec<TokenStream> =
                    (0..batch_rank).map(|i| quote! { shape[#i] }).collect();
                quote! {
                    let #output = {
                        let shape = #input.shape();
                        let batch_size: usize = shape[..#batch_rank].iter().product();
                        let m = shape[#batch_rank];
                        let flat = #input.reshape([batch_size, m, m]);
                        let mut dets: alloc::vec::Vec<Tensor<B, 1, Float>> = alloc::vec::Vec::with_capacity(batch_size);
                        for i in 0..batch_size {
                            let matrix: Tensor<B, 2, Float> =
                                flat.clone().slice([i..(i + 1), 0..m, 0..m]).squeeze_dims::<2>(&[0]);
                            let (lu_mat, permutations) =
                                burn::tensor::linalg::lu_decomposition(matrix);
                            let det_u =
                                burn::tensor::linalg::diag::<B, 2usize, 1usize, Float>(lu_mat)
                                    .prod();
                            let n = permutations.dims()[0];
                            let perm_f = permutations.float().cast(burn::tensor::DType::F32);
                            let perm_col = perm_f.clone().reshape([n, 1]);
                            let perm_row = perm_f.reshape([1, n]);
                            let inv_count = (perm_row - perm_col)
                                .triu(1i64)
                                .lower_elem(0.0f32)
                                .int()
                                .cast(burn::tensor::DType::I64)
                                .sum();
                            let parity = inv_count.remainder_scalar(2i64);
                            let device = det_u.device();
                            let dtype = det_u.dtype();
                            let sign = Tensor::<B, 1, Float>::from_data(
                                [1.0f32, -1.0f32],
                                (&device, dtype,)
                            )
                            .select(0, parity);
                            dets.push(det_u * sign);
                        }
                        let flat_out = Tensor::cat(dets, 0);
                        flat_out.reshape([#(#batch_dims),*])
                    };
                }
            }
        }
    }
}

#[cfg(test)]
mod tests {
    use super::super::test_helpers::*;
    use burn::tensor::DType;
    use insta::assert_snapshot;
    use onnx_ir::node::det::DetNodeBuilder;

    #[test]
    fn test_det_2d_forward() {
        let node = DetNodeBuilder::new("det1")
            .input_tensor("input", 2, DType::F32)
            .output_scalar_tensor("output", DType::F32)
            .build();
        let code = codegen_forward_default(&node);
        assert_snapshot!(code, @r"
        pub fn forward(&self, input: Tensor<B, 2>) -> Tensor<B, 1> {
            let output = {
                let (lu_mat, permutations) = burn::tensor::linalg::lu_decomposition(input);
                let det_u = burn::tensor::linalg::diag::<B, 2usize, 1usize, Float>(lu_mat)
                    .prod();
                let n = permutations.dims()[0];
                let perm_f = permutations.float().cast(burn::tensor::DType::F32);
                let perm_col = perm_f.clone().reshape([n, 1]);
                let perm_row = perm_f.reshape([1, n]);
                let inv_count = (perm_row - perm_col)
                    .triu(1i64)
                    .lower_elem(0.0f32)
                    .int()
                    .cast(burn::tensor::DType::I64)
                    .sum();
                let parity = inv_count.remainder_scalar(2i64);
                let device = det_u.device();
                let dtype = det_u.dtype();
                let sign = Tensor::<B, 1, Float>::from_data([1.0f32, -1.0f32], (&device, dtype))
                    .select(0, parity);
                det_u * sign
            };
            output
        }
        ");
    }

    #[test]
    fn test_det_3d_forward() {
        let node = DetNodeBuilder::new("det1")
            .input_tensor("input", 3, DType::F32)
            .output_tensor("output", 1, DType::F32)
            .build();
        let code = codegen_forward_default(&node);
        assert_snapshot!(code, @r"
        pub fn forward(&self, input: Tensor<B, 3>) -> Tensor<B, 1> {
            let output = {
                let shape = input.shape();
                let batch_size: usize = shape[..1usize].iter().product();
                let m = shape[1usize];
                let flat = input.reshape([batch_size, m, m]);
                let mut dets: alloc::vec::Vec<Tensor<B, 1, Float>> = alloc::vec::Vec::with_capacity(
                    batch_size,
                );
                for i in 0..batch_size {
                    let matrix: Tensor<B, 2, Float> = flat
                        .clone()
                        .slice([i..(i + 1), 0..m, 0..m])
                        .squeeze_dims::<2>(&[0]);
                    let (lu_mat, permutations) = burn::tensor::linalg::lu_decomposition(matrix);
                    let det_u = burn::tensor::linalg::diag::<B, 2usize, 1usize, Float>(lu_mat)
                        .prod();
                    let n = permutations.dims()[0];
                    let perm_f = permutations.float().cast(burn::tensor::DType::F32);
                    let perm_col = perm_f.clone().reshape([n, 1]);
                    let perm_row = perm_f.reshape([1, n]);
                    let inv_count = (perm_row - perm_col)
                        .triu(1i64)
                        .lower_elem(0.0f32)
                        .int()
                        .cast(burn::tensor::DType::I64)
                        .sum();
                    let parity = inv_count.remainder_scalar(2i64);
                    let device = det_u.device();
                    let dtype = det_u.dtype();
                    let sign = Tensor::<
                        B,
                        1,
                        Float,
                    >::from_data([1.0f32, -1.0f32], (&device, dtype))
                        .select(0, parity);
                    dets.push(det_u * sign);
                }
                let flat_out = Tensor::cat(dets, 0);
                flat_out.reshape([shape[0usize]])
            };
            output
        }
        ");
    }
}