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

use super::broadcast_helpers::align_rhs_for_lhs_rank;
use super::prelude::*;

impl NodeCodegen for onnx_ir::dequantize_linear::DequantizeLinearNode {
    fn inputs(&self) -> &[Argument] {
        &self.inputs
    }

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

    fn forward(&self, scope: &mut ScopeAtPosition<'_>) -> TokenStream {
        let x_arg = self.inputs.first().unwrap();
        let scale_arg = self.inputs.get(1).unwrap();
        let output = arg_to_ident(self.outputs.first().unwrap());

        let x = scope.arg(x_arg);
        let scale = scope.arg(scale_arg);

        let out_dtype = self.outputs.first().unwrap().ty.elem_type();
        let out_dtype_tokens = out_dtype.to_tokens();

        let x_expr = if x_arg.ty.elem_type() == out_dtype {
            quote! { #x }
        } else if x_arg.ty.elem_type().is_float() {
            quote! { (#x).cast(#out_dtype_tokens) }
        } else if x_arg.ty.elem_type().is_uint() {
            quote! { (#x).cast(burn::tensor::DType::I32).float().cast(#out_dtype_tokens) }
        } else {
            quote! { (#x).float().cast(#out_dtype_tokens) }
        };

        let scale_expr = if scale_arg.ty.elem_type() == out_dtype {
            quote! { #scale }
        } else if scale_arg.ty.elem_type().is_float() {
            quote! { (#scale).cast(#out_dtype_tokens) }
        } else {
            quote! { (#scale).float().cast(#out_dtype_tokens) }
        };

        let x_rank = x_arg.ty.rank();
        let scale_rank = scale_arg.ty.rank();
        let scale_expr = align_rhs_for_lhs_rank(scale_expr, x_rank, scale_rank, self.config.axis);

        let centered_expr = if let Some(zp_arg) = self.inputs.get(2) {
            let zp = scope.arg(zp_arg);
            let zp_expr = if zp_arg.ty.elem_type() == out_dtype {
                quote! { #zp }
            } else if zp_arg.ty.elem_type().is_float() {
                quote! { (#zp).cast(#out_dtype_tokens) }
            } else if zp_arg.ty.elem_type().is_uint() {
                quote! { (#zp).cast(burn::tensor::DType::I32).float().cast(#out_dtype_tokens) }
            } else {
                quote! { (#zp).float().cast(#out_dtype_tokens) }
            };
            let zp_expr =
                align_rhs_for_lhs_rank(zp_expr, x_rank, zp_arg.ty.rank(), self.config.axis);
            quote! { (#x_expr).sub(#zp_expr) }
        } else {
            x_expr
        };

        quote! {
            let #output = (#centered_expr).mul(#scale_expr);
        }
    }
}

#[cfg(test)]
mod tests {
    use super::super::test_helpers::*;
    use burn::tensor::DType;
    use insta::assert_snapshot;
    use onnx_ir::dequantize_linear::{
        DequantizeLinearConfig, DequantizeLinearNode, DequantizeLinearNodeBuilder,
    };

    fn create_node(name: &str, with_zero_point: bool) -> DequantizeLinearNode {
        let mut builder = DequantizeLinearNodeBuilder::new(name)
            .input_tensor("x", 2, DType::U8)
            .input_tensor("x_scale", 0, DType::F32)
            .output_tensor("y", 2, DType::F32)
            .config(DequantizeLinearConfig::default());

        if with_zero_point {
            builder = builder.input_tensor("x_zero_point", 0, DType::U8);
        }

        builder.build()
    }

    #[test]
    fn test_dequantize_linear_forward_without_zero_point() {
        let node = create_node("dq", false);
        let code = codegen_forward_default(&node);
        assert_snapshot!(code, @r"
        pub fn forward(&self, x: Tensor<B, 2, Int>, x_scale: Tensor<B, 0>) -> Tensor<B, 2> {
            let y = ((x).cast(burn::tensor::DType::I32).float().cast(burn::tensor::DType::F32))
                .mul((x_scale).unsqueeze_dims(&[0isize, 1isize]));
            y
        }
        ");
    }

    #[test]
    fn test_dequantize_linear_forward_with_zero_point() {
        let node = create_node("dq", true);
        let code = codegen_forward_default(&node);
        assert_snapshot!(code, @r"
        pub fn forward(
            &self,
            x: Tensor<B, 2, Int>,
            x_scale: Tensor<B, 0>,
            x_zero_point: Tensor<B, 0, Int>,
        ) -> Tensor<B, 2> {
            let y = (((x).cast(burn::tensor::DType::I32).float().cast(burn::tensor::DType::F32))
                .sub(
                    ((x_zero_point)
                        .cast(burn::tensor::DType::I32)
                        .float()
                        .cast(burn::tensor::DType::F32))
                        .unsqueeze_dims(&[0isize, 1isize]),
                ))
                .mul((x_scale).unsqueeze_dims(&[0isize, 1isize]));
            y
        }
        ");
    }
}